net/xfrm/xfrm_policy.c

   1 /*
   2  * xfrm_policy.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      Kazunori MIYAZAWA @USAGI
  10  *      YOSHIFUJI Hideaki
  11  *              Split up af-specific portion
  12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  13  *
  14  */
  15
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/kmod.h>
  19 #include <linux/list.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/module.h>
  26 #include <linux/cache.h>
  27 #include <linux/audit.h>
  28 #include <net/dst.h>
  29 #include <net/flow.h>
  30 #include <net/xfrm.h>
  31 #include <net/ip.h>
  32 #ifdef CONFIG_XFRM_STATISTICS
  33 #include <net/snmp.h>
  34 #endif
  35
  36 #include "xfrm_hash.h"
  37
  38 DEFINE_MUTEX(xfrm_cfg_mutex);
  39 EXPORT_SYMBOL(xfrm_cfg_mutex);
  40
  41 static DEFINE_SPINLOCK(xfrm_policy_sk_bundle_lock);
  42 static struct dst_entry *xfrm_policy_sk_bundles;
  43 static DEFINE_RWLOCK(xfrm_policy_lock);
  44
  45 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
  46 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
  47
  48 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  49
  50 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
  51 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
  52 static void xfrm_init_pmtu(struct dst_entry *dst);
  53 static int stale_bundle(struct dst_entry *dst);
  54 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
  55
  56
  57 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  58                                                 int dir);
  59
  60 static inline bool
  61 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  62 {
  63         const struct flowi4 *fl4 = &fl->u.ip4;
  64
  65         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
  66                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
  67                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  68                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  69                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  70                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  71 }
  72
  73 static inline bool
  74 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  75 {
  76         const struct flowi6 *fl6 = &fl->u.ip6;
  77
  78         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  79                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  80                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  81                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  82                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  83                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  84 }
  85
  86 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  87                          unsigned short family)
  88 {
  89         switch (family) {
  90         case AF_INET:
  91                 return __xfrm4_selector_match(sel, fl);
  92         case AF_INET6:
  93                 return __xfrm6_selector_match(sel, fl);
  94         }
  95         return false;
  96 }
  97
  98 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
  99                                                   const xfrm_address_t *saddr,
 100                                                   const xfrm_address_t *daddr,
 101                                                   int family)
 102 {
 103         struct xfrm_policy_afinfo *afinfo;
 104         struct dst_entry *dst;
 105
 106         afinfo = xfrm_policy_get_afinfo(family);
 107         if (unlikely(afinfo == NULL))
 108                 return ERR_PTR(-EAFNOSUPPORT);
 109
 110         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 111
 112         xfrm_policy_put_afinfo(afinfo);
 113
 114         return dst;
 115 }
 116
 117 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 118                                                 xfrm_address_t *prev_saddr,
 119                                                 xfrm_address_t *prev_daddr,
 120                                                 int family)
 121 {
 122         struct net *net = xs_net(x);
 123         xfrm_address_t *saddr = &x->props.saddr;
 124         xfrm_address_t *daddr = &x->id.daddr;
 125         struct dst_entry *dst;
 126
 127         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 128                 saddr = x->coaddr;
 129                 daddr = prev_daddr;
 130         }
 131         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 132                 saddr = prev_saddr;
 133                 daddr = x->coaddr;
 134         }
 135
 136         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 137
 138         if (!IS_ERR(dst)) {
 139                 if (prev_saddr != saddr)
 140                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 141                 if (prev_daddr != daddr)
 142                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 143         }
 144
 145         return dst;
 146 }
 147
 148 static inline unsigned long make_jiffies(long secs)
 149 {
 150         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 151                 return MAX_SCHEDULE_TIMEOUT-1;
 152         else
 153                 return secs*HZ;
 154 }
 155
 156 static void xfrm_policy_timer(unsigned long data)
 157 {
 158         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 159         unsigned long now = get_seconds();
 160         long next = LONG_MAX;
 161         int warn = 0;
 162         int dir;
 163
 164         read_lock(&xp->lock);
 165
 166         if (unlikely(xp->walk.dead))
 167                 goto out;
 168
 169         dir = xfrm_policy_id2dir(xp->index);
 170
 171         if (xp->lft.hard_add_expires_seconds) {
 172                 long tmo = xp->lft.hard_add_expires_seconds +
 173                         xp->curlft.add_time - now;
 174                 if (tmo <= 0)
 175                         goto expired;
 176                 if (tmo < next)
 177                         next = tmo;
 178         }
 179         if (xp->lft.hard_use_expires_seconds) {
 180                 long tmo = xp->lft.hard_use_expires_seconds +
 181                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 182                 if (tmo <= 0)
 183                         goto expired;
 184                 if (tmo < next)
 185                         next = tmo;
 186         }
 187         if (xp->lft.soft_add_expires_seconds) {
 188                 long tmo = xp->lft.soft_add_expires_seconds +
 189                         xp->curlft.add_time - now;
 190                 if (tmo <= 0) {
 191                         warn = 1;
 192                         tmo = XFRM_KM_TIMEOUT;
 193                 }
 194                 if (tmo < next)
 195                         next = tmo;
 196         }
 197         if (xp->lft.soft_use_expires_seconds) {
 198                 long tmo = xp->lft.soft_use_expires_seconds +
 199                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 200                 if (tmo <= 0) {
 201                         warn = 1;
 202                         tmo = XFRM_KM_TIMEOUT;
 203                 }
 204                 if (tmo < next)
 205                         next = tmo;
 206         }
 207
 208         if (warn)
 209                 km_policy_expired(xp, dir, 0, 0);
 210         if (next != LONG_MAX &&
 211             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 212                 xfrm_pol_hold(xp);
 213
 214 out:
 215         read_unlock(&xp->lock);
 216         xfrm_pol_put(xp);
 217         return;
 218
 219 expired:
 220         read_unlock(&xp->lock);
 221         if (!xfrm_policy_delete(xp, dir))
 222                 km_policy_expired(xp, dir, 1, 0);
 223         xfrm_pol_put(xp);
 224 }
 225
 226 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 227 {
 228         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 229
 230         if (unlikely(pol->walk.dead))
 231                 flo = NULL;
 232         else
 233                 xfrm_pol_hold(pol);
 234
 235         return flo;
 236 }
 237
 238 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 239 {
 240         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 241
 242         return !pol->walk.dead;
 243 }
 244
 245 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 246 {
 247         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 248 }
 249
 250 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 251         .get = xfrm_policy_flo_get,
 252         .check = xfrm_policy_flo_check,
 253         .delete = xfrm_policy_flo_delete,
 254 };
 255
 256 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 257  * SPD calls.
 258  */
 259
 260 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 261 {
 262         struct xfrm_policy *policy;
 263
 264         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 265
 266         if (policy) {
 267                 write_pnet(&policy->xp_net, net);
 268                 INIT_LIST_HEAD(&policy->walk.all);
 269                 INIT_HLIST_NODE(&policy->bydst);
 270                 INIT_HLIST_NODE(&policy->byidx);
 271                 rwlock_init(&policy->lock);
 272                 atomic_set(&policy->refcnt, 1);
 273                 setup_timer(&policy->timer, xfrm_policy_timer,
 274                                 (unsigned long)policy);
 275                 policy->flo.ops = &xfrm_policy_fc_ops;
 276         }
 277         return policy;
 278 }
 279 EXPORT_SYMBOL(xfrm_policy_alloc);
 280
 281 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 282
 283 void xfrm_policy_destroy(struct xfrm_policy *policy)
 284 {
 285         BUG_ON(!policy->walk.dead);
 286
 287         if (del_timer(&policy->timer))
 288                 BUG();
 289
 290         security_xfrm_policy_free(policy->security);
 291         kfree(policy);
 292 }
 293 EXPORT_SYMBOL(xfrm_policy_destroy);
 294
 295 /* Rule must be locked. Release descentant resources, announce
 296  * entry dead. The rule must be unlinked from lists to the moment.
 297  */
 298
 299 static void xfrm_policy_kill(struct xfrm_policy *policy)
 300 {
 301         policy->walk.dead = 1;
 302
 303         atomic_inc(&policy->genid);
 304
 305         if (del_timer(&policy->timer))
 306                 xfrm_pol_put(policy);
 307
 308         xfrm_pol_put(policy);
 309 }
 310
 311 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 312
 313 static inline unsigned int idx_hash(struct net *net, u32 index)
 314 {
 315         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 316 }
 317
 318 static struct hlist_head *policy_hash_bysel(struct net *net,
 319                                             const struct xfrm_selector *sel,
 320                                             unsigned short family, int dir)
 321 {
 322         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 323         unsigned int hash = __sel_hash(sel, family, hmask);
 324
 325         return (hash == hmask + 1 ?
 326                 &net->xfrm.policy_inexact[dir] :
 327                 net->xfrm.policy_bydst[dir].table + hash);
 328 }
 329
 330 static struct hlist_head *policy_hash_direct(struct net *net,
 331                                              const xfrm_address_t *daddr,
 332                                              const xfrm_address_t *saddr,
 333                                              unsigned short family, int dir)
 334 {
 335         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 336         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 337
 338         return net->xfrm.policy_bydst[dir].table + hash;
 339 }
 340
 341 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 342                                    struct hlist_head *ndsttable,
 343                                    unsigned int nhashmask)
 344 {
 345         struct hlist_node *entry, *tmp, *entry0 = NULL;
 346         struct xfrm_policy *pol;
 347         unsigned int h0 = 0;
 348
 349 redo:
 350         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
 351                 unsigned int h;
 352
 353                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 354                                 pol->family, nhashmask);
 355                 if (!entry0) {
 356                         hlist_del(entry);
 357                         hlist_add_head(&pol->bydst, ndsttable+h);
 358                         h0 = h;
 359                 } else {
 360                         if (h != h0)
 361                                 continue;
 362                         hlist_del(entry);
 363                         hlist_add_after(entry0, &pol->bydst);
 364                 }
 365                 entry0 = entry;
 366         }
 367         if (!hlist_empty(list)) {
 368                 entry0 = NULL;
 369                 goto redo;
 370         }
 371 }
 372
 373 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 374                                    struct hlist_head *nidxtable,
 375                                    unsigned int nhashmask)
 376 {
 377         struct hlist_node *entry, *tmp;
 378         struct xfrm_policy *pol;
 379
 380         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
 381                 unsigned int h;
 382
 383                 h = __idx_hash(pol->index, nhashmask);
 384                 hlist_add_head(&pol->byidx, nidxtable+h);
 385         }
 386 }
 387
 388 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 389 {
 390         return ((old_hmask + 1) << 1) - 1;
 391 }
 392
 393 static void xfrm_bydst_resize(struct net *net, int dir)
 394 {
 395         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 396         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 397         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 398         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 399         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 400         int i;
 401
 402         if (!ndst)
 403                 return;
 404
 405         write_lock_bh(&xfrm_policy_lock);
 406
 407         for (i = hmask; i >= 0; i--)
 408                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 409
 410         net->xfrm.policy_bydst[dir].table = ndst;
 411         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 412
 413         write_unlock_bh(&xfrm_policy_lock);
 414
 415         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 416 }
 417
 418 static void xfrm_byidx_resize(struct net *net, int total)
 419 {
 420         unsigned int hmask = net->xfrm.policy_idx_hmask;
 421         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 422         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 423         struct hlist_head *oidx = net->xfrm.policy_byidx;
 424         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 425         int i;
 426
 427         if (!nidx)
 428                 return;
 429
 430         write_lock_bh(&xfrm_policy_lock);
 431
 432         for (i = hmask; i >= 0; i--)
 433                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 434
 435         net->xfrm.policy_byidx = nidx;
 436         net->xfrm.policy_idx_hmask = nhashmask;
 437
 438         write_unlock_bh(&xfrm_policy_lock);
 439
 440         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 441 }
 442
 443 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 444 {
 445         unsigned int cnt = net->xfrm.policy_count[dir];
 446         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 447
 448         if (total)
 449                 *total += cnt;
 450
 451         if ((hmask + 1) < xfrm_policy_hashmax &&
 452             cnt > hmask)
 453                 return 1;
 454
 455         return 0;
 456 }
 457
 458 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 459 {
 460         unsigned int hmask = net->xfrm.policy_idx_hmask;
 461
 462         if ((hmask + 1) < xfrm_policy_hashmax &&
 463             total > hmask)
 464                 return 1;
 465
 466         return 0;
 467 }
 468
 469 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 470 {
 471         read_lock_bh(&xfrm_policy_lock);
 472         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 473         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 474         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 475         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 476         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 477         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 478         si->spdhcnt = net->xfrm.policy_idx_hmask;
 479         si->spdhmcnt = xfrm_policy_hashmax;
 480         read_unlock_bh(&xfrm_policy_lock);
 481 }
 482 EXPORT_SYMBOL(xfrm_spd_getinfo);
 483
 484 static DEFINE_MUTEX(hash_resize_mutex);
 485 static void xfrm_hash_resize(struct work_struct *work)
 486 {
 487         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 488         int dir, total;
 489
 490         mutex_lock(&hash_resize_mutex);
 491
 492         total = 0;
 493         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 494                 if (xfrm_bydst_should_resize(net, dir, &total))
 495                         xfrm_bydst_resize(net, dir);
 496         }
 497         if (xfrm_byidx_should_resize(net, total))
 498                 xfrm_byidx_resize(net, total);
 499
 500         mutex_unlock(&hash_resize_mutex);
 501 }
 502
 503 /* Generate new index... KAME seems to generate them ordered by cost
 504  * of an absolute inpredictability of ordering of rules. This will not pass. */
 505 static u32 xfrm_gen_index(struct net *net, int dir)
 506 {
 507         static u32 idx_generator;
 508
 509         for (;;) {
 510                 struct hlist_node *entry;
 511                 struct hlist_head *list;
 512                 struct xfrm_policy *p;
 513                 u32 idx;
 514                 int found;
 515
 516                 idx = (idx_generator | dir);
 517                 idx_generator += 8;
 518                 if (idx == 0)
 519                         idx = 8;
 520                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 521                 found = 0;
 522                 hlist_for_each_entry(p, entry, list, byidx) {
 523                         if (p->index == idx) {
 524                                 found = 1;
 525                                 break;
 526                         }
 527                 }
 528                 if (!found)
 529                         return idx;
 530         }
 531 }
 532
 533 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 534 {
 535         u32 *p1 = (u32 *) s1;
 536         u32 *p2 = (u32 *) s2;
 537         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 538         int i;
 539
 540         for (i = 0; i < len; i++) {
 541                 if (p1[i] != p2[i])
 542                         return 1;
 543         }
 544
 545         return 0;
 546 }
 547
 548 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 549 {
 550         struct net *net = xp_net(policy);
 551         struct xfrm_policy *pol;
 552         struct xfrm_policy *delpol;
 553         struct hlist_head *chain;
 554         struct hlist_node *entry, *newpos;
 555         u32 mark = policy->mark.v & policy->mark.m;
 556
 557         write_lock_bh(&xfrm_policy_lock);
 558         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 559         delpol = NULL;
 560         newpos = NULL;
 561         hlist_for_each_entry(pol, entry, chain, bydst) {
 562                 if (pol->type == policy->type &&
 563                     !selector_cmp(&pol->selector, &policy->selector) &&
 564                     (mark & pol->mark.m) == pol->mark.v &&
 565                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 566                     !WARN_ON(delpol)) {
 567                         if (excl) {
 568                                 write_unlock_bh(&xfrm_policy_lock);
 569                                 return -EEXIST;
 570                         }
 571                         delpol = pol;
 572                         if (policy->priority > pol->priority)
 573                                 continue;
 574                 } else if (policy->priority >= pol->priority) {
 575                         newpos = &pol->bydst;
 576                         continue;
 577                 }
 578                 if (delpol)
 579                         break;
 580         }
 581         if (newpos)
 582                 hlist_add_after(newpos, &policy->bydst);
 583         else
 584                 hlist_add_head(&policy->bydst, chain);
 585         xfrm_pol_hold(policy);
 586         net->xfrm.policy_count[dir]++;
 587         atomic_inc(&flow_cache_genid);
 588         if (delpol)
 589                 __xfrm_policy_unlink(delpol, dir);
 590         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
 591         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 592         policy->curlft.add_time = get_seconds();
 593         policy->curlft.use_time = 0;
 594         if (!mod_timer(&policy->timer, jiffies + HZ))
 595                 xfrm_pol_hold(policy);
 596         list_add(&policy->walk.all, &net->xfrm.policy_all);
 597         write_unlock_bh(&xfrm_policy_lock);
 598
 599         if (delpol)
 600                 xfrm_policy_kill(delpol);
 601         else if (xfrm_bydst_should_resize(net, dir, NULL))
 602                 schedule_work(&net->xfrm.policy_hash_work);
 603
 604         return 0;
 605 }
 606 EXPORT_SYMBOL(xfrm_policy_insert);
 607
 608 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 609                                           int dir, struct xfrm_selector *sel,
 610                                           struct xfrm_sec_ctx *ctx, int delete,
 611                                           int *err)
 612 {
 613         struct xfrm_policy *pol, *ret;
 614         struct hlist_head *chain;
 615         struct hlist_node *entry;
 616
 617         *err = 0;
 618         write_lock_bh(&xfrm_policy_lock);
 619         chain = policy_hash_bysel(net, sel, sel->family, dir);
 620         ret = NULL;
 621         hlist_for_each_entry(pol, entry, chain, bydst) {
 622                 if (pol->type == type &&
 623                     (mark & pol->mark.m) == pol->mark.v &&
 624                     !selector_cmp(sel, &pol->selector) &&
 625                     xfrm_sec_ctx_match(ctx, pol->security)) {
 626                         xfrm_pol_hold(pol);
 627                         if (delete) {
 628                                 *err = security_xfrm_policy_delete(
 629                                                                 pol->security);
 630                                 if (*err) {
 631                                         write_unlock_bh(&xfrm_policy_lock);
 632                                         return pol;
 633                                 }
 634                                 __xfrm_policy_unlink(pol, dir);
 635                         }
 636                         ret = pol;
 637                         break;
 638                 }
 639         }
 640         write_unlock_bh(&xfrm_policy_lock);
 641
 642         if (ret && delete)
 643                 xfrm_policy_kill(ret);
 644         return ret;
 645 }
 646 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 647
 648 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 649                                      int dir, u32 id, int delete, int *err)
 650 {
 651         struct xfrm_policy *pol, *ret;
 652         struct hlist_head *chain;
 653         struct hlist_node *entry;
 654
 655         *err = -ENOENT;
 656         if (xfrm_policy_id2dir(id) != dir)
 657                 return NULL;
 658
 659         *err = 0;
 660         write_lock_bh(&xfrm_policy_lock);
 661         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 662         ret = NULL;
 663         hlist_for_each_entry(pol, entry, chain, byidx) {
 664                 if (pol->type == type && pol->index == id &&
 665                     (mark & pol->mark.m) == pol->mark.v) {
 666                         xfrm_pol_hold(pol);
 667                         if (delete) {
 668                                 *err = security_xfrm_policy_delete(
 669                                                                 pol->security);
 670                                 if (*err) {
 671                                         write_unlock_bh(&xfrm_policy_lock);
 672                                         return pol;
 673                                 }
 674                                 __xfrm_policy_unlink(pol, dir);
 675                         }
 676                         ret = pol;
 677                         break;
 678                 }
 679         }
 680         write_unlock_bh(&xfrm_policy_lock);
 681
 682         if (ret && delete)
 683                 xfrm_policy_kill(ret);
 684         return ret;
 685 }
 686 EXPORT_SYMBOL(xfrm_policy_byid);
 687
 688 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 689 static inline int
 690 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 691 {
 692         int dir, err = 0;
 693
 694         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 695                 struct xfrm_policy *pol;
 696                 struct hlist_node *entry;
 697                 int i;
 698
 699                 hlist_for_each_entry(pol, entry,
 700                                      &net->xfrm.policy_inexact[dir], bydst) {
 701                         if (pol->type != type)
 702                                 continue;
 703                         err = security_xfrm_policy_delete(pol->security);
 704                         if (err) {
 705                                 xfrm_audit_policy_delete(pol, 0,
 706                                                          audit_info->loginuid,
 707                                                          audit_info->sessionid,
 708                                                          audit_info->secid);
 709                                 return err;
 710                         }
 711                 }
 712                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 713                         hlist_for_each_entry(pol, entry,
 714                                              net->xfrm.policy_bydst[dir].table + i,
 715                                              bydst) {
 716                                 if (pol->type != type)
 717                                         continue;
 718                                 err = security_xfrm_policy_delete(
 719                                                                 pol->security);
 720                                 if (err) {
 721                                         xfrm_audit_policy_delete(pol, 0,
 722                                                         audit_info->loginuid,
 723                                                         audit_info->sessionid,
 724                                                         audit_info->secid);
 725                                         return err;
 726                                 }
 727                         }
 728                 }
 729         }
 730         return err;
 731 }
 732 #else
 733 static inline int
 734 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 735 {
 736         return 0;
 737 }
 738 #endif
 739
 740 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
 741 {
 742         int dir, err = 0, cnt = 0;
 743
 744         write_lock_bh(&xfrm_policy_lock);
 745
 746         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
 747         if (err)
 748                 goto out;
 749
 750         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 751                 struct xfrm_policy *pol;
 752                 struct hlist_node *entry;
 753                 int i;
 754
 755         again1:
 756                 hlist_for_each_entry(pol, entry,
 757                                      &net->xfrm.policy_inexact[dir], bydst) {
 758                         if (pol->type != type)
 759                                 continue;
 760                         __xfrm_policy_unlink(pol, dir);
 761                         write_unlock_bh(&xfrm_policy_lock);
 762                         cnt++;
 763
 764                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 765                                                  audit_info->sessionid,
 766                                                  audit_info->secid);
 767
 768                         xfrm_policy_kill(pol);
 769
 770                         write_lock_bh(&xfrm_policy_lock);
 771                         goto again1;
 772                 }
 773
 774                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 775         again2:
 776                         hlist_for_each_entry(pol, entry,
 777                                              net->xfrm.policy_bydst[dir].table + i,
 778                                              bydst) {
 779                                 if (pol->type != type)
 780                                         continue;
 781                                 __xfrm_policy_unlink(pol, dir);
 782                                 write_unlock_bh(&xfrm_policy_lock);
 783                                 cnt++;
 784
 785                                 xfrm_audit_policy_delete(pol, 1,
 786                                                          audit_info->loginuid,
 787                                                          audit_info->sessionid,
 788                                                          audit_info->secid);
 789                                 xfrm_policy_kill(pol);
 790
 791                                 write_lock_bh(&xfrm_policy_lock);
 792                                 goto again2;
 793                         }
 794                 }
 795
 796         }
 797         if (!cnt)
 798                 err = -ESRCH;
 799 out:
 800         write_unlock_bh(&xfrm_policy_lock);
 801         return err;
 802 }
 803 EXPORT_SYMBOL(xfrm_policy_flush);
 804
 805 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 806                      int (*func)(struct xfrm_policy *, int, int, void*),
 807                      void *data)
 808 {
 809         struct xfrm_policy *pol;
 810         struct xfrm_policy_walk_entry *x;
 811         int error = 0;
 812
 813         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 814             walk->type != XFRM_POLICY_TYPE_ANY)
 815                 return -EINVAL;
 816
 817         if (list_empty(&walk->walk.all) && walk->seq != 0)
 818                 return 0;
 819
 820         write_lock_bh(&xfrm_policy_lock);
 821         if (list_empty(&walk->walk.all))
 822                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 823         else
 824                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 825         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 826                 if (x->dead)
 827                         continue;
 828                 pol = container_of(x, struct xfrm_policy, walk);
 829                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 830                     walk->type != pol->type)
 831                         continue;
 832                 error = func(pol, xfrm_policy_id2dir(pol->index),
 833                              walk->seq, data);
 834                 if (error) {
 835                         list_move_tail(&walk->walk.all, &x->all);
 836                         goto out;
 837                 }
 838                 walk->seq++;
 839         }
 840         if (walk->seq == 0) {
 841                 error = -ENOENT;
 842                 goto out;
 843         }
 844         list_del_init(&walk->walk.all);
 845 out:
 846         write_unlock_bh(&xfrm_policy_lock);
 847         return error;
 848 }
 849 EXPORT_SYMBOL(xfrm_policy_walk);
 850
 851 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 852 {
 853         INIT_LIST_HEAD(&walk->walk.all);
 854         walk->walk.dead = 1;
 855         walk->type = type;
 856         walk->seq = 0;
 857 }
 858 EXPORT_SYMBOL(xfrm_policy_walk_init);
 859
 860 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk)
 861 {
 862         if (list_empty(&walk->walk.all))
 863                 return;
 864
 865         write_lock_bh(&xfrm_policy_lock);
 866         list_del(&walk->walk.all);
 867         write_unlock_bh(&xfrm_policy_lock);
 868 }
 869 EXPORT_SYMBOL(xfrm_policy_walk_done);
 870
 871 /*
 872  * Find policy to apply to this flow.
 873  *
 874  * Returns 0 if policy found, else an -errno.
 875  */
 876 static int xfrm_policy_match(const struct xfrm_policy *pol,
 877                              const struct flowi *fl,
 878                              u8 type, u16 family, int dir)
 879 {
 880         const struct xfrm_selector *sel = &pol->selector;
 881         int ret = -ESRCH;
 882         bool match;
 883
 884         if (pol->family != family ||
 885             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
 886             pol->type != type)
 887                 return ret;
 888
 889         match = xfrm_selector_match(sel, fl, family);
 890         if (match)
 891                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
 892                                                   dir);
 893
 894         return ret;
 895 }
 896
 897 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 898                                                      const struct flowi *fl,
 899                                                      u16 family, u8 dir)
 900 {
 901         int err;
 902         struct xfrm_policy *pol, *ret;
 903         const xfrm_address_t *daddr, *saddr;
 904         struct hlist_node *entry;
 905         struct hlist_head *chain;
 906         u32 priority = ~0U;
 907
 908         daddr = xfrm_flowi_daddr(fl, family);
 909         saddr = xfrm_flowi_saddr(fl, family);
 910         if (unlikely(!daddr || !saddr))
 911                 return NULL;
 912
 913         read_lock_bh(&xfrm_policy_lock);
 914         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 915         ret = NULL;
 916         hlist_for_each_entry(pol, entry, chain, bydst) {
 917                 err = xfrm_policy_match(pol, fl, type, family, dir);
 918                 if (err) {
 919                         if (err == -ESRCH)
 920                                 continue;
 921                         else {
 922                                 ret = ERR_PTR(err);
 923                                 goto fail;
 924                         }
 925                 } else {
 926                         ret = pol;
 927                         priority = ret->priority;
 928                         break;
 929                 }
 930         }
 931         chain = &net->xfrm.policy_inexact[dir];
 932         hlist_for_each_entry(pol, entry, chain, bydst) {
 933                 err = xfrm_policy_match(pol, fl, type, family, dir);
 934                 if (err) {
 935                         if (err == -ESRCH)
 936                                 continue;
 937                         else {
 938                                 ret = ERR_PTR(err);
 939                                 goto fail;
 940                         }
 941                 } else if (pol->priority < priority) {
 942                         ret = pol;
 943                         break;
 944                 }
 945         }
 946         if (ret)
 947                 xfrm_pol_hold(ret);
 948 fail:
 949         read_unlock_bh(&xfrm_policy_lock);
 950
 951         return ret;
 952 }
 953
 954 static struct xfrm_policy *
 955 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
 956 {
 957 #ifdef CONFIG_XFRM_SUB_POLICY
 958         struct xfrm_policy *pol;
 959
 960         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
 961         if (pol != NULL)
 962                 return pol;
 963 #endif
 964         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
 965 }
 966
 967 static struct flow_cache_object *
 968 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
 969                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
 970 {
 971         struct xfrm_policy *pol;
 972
 973         if (old_obj)
 974                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
 975
 976         pol = __xfrm_policy_lookup(net, fl, family, dir);
 977         if (IS_ERR_OR_NULL(pol))
 978                 return ERR_CAST(pol);
 979
 980         /* Resolver returns two references:
 981          * one for cache and one for caller of flow_cache_lookup() */
 982         xfrm_pol_hold(pol);
 983
 984         return &pol->flo;
 985 }
 986
 987 static inline int policy_to_flow_dir(int dir)
 988 {
 989         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 990             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 991             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 992                 return dir;
 993         switch (dir) {
 994         default:
 995         case XFRM_POLICY_IN:
 996                 return FLOW_DIR_IN;
 997         case XFRM_POLICY_OUT:
 998                 return FLOW_DIR_OUT;
 999         case XFRM_POLICY_FWD:
1000                 return FLOW_DIR_FWD;
1001         }
1002 }
1003
1004 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1005                                                  const struct flowi *fl)
1006 {
1007         struct xfrm_policy *pol;
1008
1009         read_lock_bh(&xfrm_policy_lock);
1010         if ((pol = sk->sk_policy[dir]) != NULL) {
1011                 bool match = xfrm_selector_match(&pol->selector, fl,
1012                                                  sk->sk_family);
1013                 int err = 0;
1014
1015                 if (match) {
1016                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1017                                 pol = NULL;
1018                                 goto out;
1019                         }
1020                         err = security_xfrm_policy_lookup(pol->security,
1021                                                       fl->flowi_secid,
1022                                                       policy_to_flow_dir(dir));
1023                         if (!err)
1024                                 xfrm_pol_hold(pol);
1025                         else if (err == -ESRCH)
1026                                 pol = NULL;
1027                         else
1028                                 pol = ERR_PTR(err);
1029                 } else
1030                         pol = NULL;
1031         }
1032 out:
1033         read_unlock_bh(&xfrm_policy_lock);
1034         return pol;
1035 }
1036
1037 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1038 {
1039         struct net *net = xp_net(pol);
1040         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1041                                                      pol->family, dir);
1042
1043         list_add(&pol->walk.all, &net->xfrm.policy_all);
1044         hlist_add_head(&pol->bydst, chain);
1045         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1046         net->xfrm.policy_count[dir]++;
1047         xfrm_pol_hold(pol);
1048
1049         if (xfrm_bydst_should_resize(net, dir, NULL))
1050                 schedule_work(&net->xfrm.policy_hash_work);
1051 }
1052
1053 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1054                                                 int dir)
1055 {
1056         struct net *net = xp_net(pol);
1057
1058         if (hlist_unhashed(&pol->bydst))
1059                 return NULL;
1060
1061         hlist_del(&pol->bydst);
1062         hlist_del(&pol->byidx);
1063         list_del(&pol->walk.all);
1064         net->xfrm.policy_count[dir]--;
1065
1066         return pol;
1067 }
1068
1069 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1070 {
1071         write_lock_bh(&xfrm_policy_lock);
1072         pol = __xfrm_policy_unlink(pol, dir);
1073         write_unlock_bh(&xfrm_policy_lock);
1074         if (pol) {
1075                 xfrm_policy_kill(pol);
1076                 return 0;
1077         }
1078         return -ENOENT;
1079 }
1080 EXPORT_SYMBOL(xfrm_policy_delete);
1081
1082 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1083 {
1084         struct net *net = xp_net(pol);
1085         struct xfrm_policy *old_pol;
1086
1087 #ifdef CONFIG_XFRM_SUB_POLICY
1088         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1089                 return -EINVAL;
1090 #endif
1091
1092         write_lock_bh(&xfrm_policy_lock);
1093         old_pol = sk->sk_policy[dir];
1094         sk->sk_policy[dir] = pol;
1095         if (pol) {
1096                 pol->curlft.add_time = get_seconds();
1097                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1098                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1099         }
1100         if (old_pol)
1101                 /* Unlinking succeeds always. This is the only function
1102                  * allowed to delete or replace socket policy.
1103                  */
1104                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1105         write_unlock_bh(&xfrm_policy_lock);
1106
1107         if (old_pol) {
1108                 xfrm_policy_kill(old_pol);
1109         }
1110         return 0;
1111 }
1112
1113 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1114 {
1115         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1116
1117         if (newp) {
1118                 newp->selector = old->selector;
1119                 if (security_xfrm_policy_clone(old->security,
1120                                                &newp->security)) {
1121                         kfree(newp);
1122                         return NULL;  /* ENOMEM */
1123                 }
1124                 newp->lft = old->lft;
1125                 newp->curlft = old->curlft;
1126                 newp->mark = old->mark;
1127                 newp->action = old->action;
1128                 newp->flags = old->flags;
1129                 newp->xfrm_nr = old->xfrm_nr;
1130                 newp->index = old->index;
1131                 newp->type = old->type;
1132                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1133                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1134                 write_lock_bh(&xfrm_policy_lock);
1135                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1136                 write_unlock_bh(&xfrm_policy_lock);
1137                 xfrm_pol_put(newp);
1138         }
1139         return newp;
1140 }
1141
1142 int __xfrm_sk_clone_policy(struct sock *sk)
1143 {
1144         struct xfrm_policy *p0 = sk->sk_policy[0],
1145                            *p1 = sk->sk_policy[1];
1146
1147         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1148         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1149                 return -ENOMEM;
1150         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1151                 return -ENOMEM;
1152         return 0;
1153 }
1154
1155 static int
1156 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1157                unsigned short family)
1158 {
1159         int err;
1160         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1161
1162         if (unlikely(afinfo == NULL))
1163                 return -EINVAL;
1164         err = afinfo->get_saddr(net, local, remote);
1165         xfrm_policy_put_afinfo(afinfo);
1166         return err;
1167 }
1168
1169 /* Resolve list of templates for the flow, given policy. */
1170
1171 static int
1172 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1173                       struct xfrm_state **xfrm, unsigned short family)
1174 {
1175         struct net *net = xp_net(policy);
1176         int nx;
1177         int i, error;
1178         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1179         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1180         xfrm_address_t tmp;
1181
1182         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1183                 struct xfrm_state *x;
1184                 xfrm_address_t *remote = daddr;
1185                 xfrm_address_t *local  = saddr;
1186                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1187
1188                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1189                     tmpl->mode == XFRM_MODE_BEET) {
1190                         remote = &tmpl->id.daddr;
1191                         local = &tmpl->saddr;
1192                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1193                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1194                                 if (error)
1195                                         goto fail;
1196                                 local = &tmp;
1197                         }
1198                 }
1199
1200                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1201
1202                 if (x && x->km.state == XFRM_STATE_VALID) {
1203                         xfrm[nx++] = x;
1204                         daddr = remote;
1205                         saddr = local;
1206                         continue;
1207                 }
1208                 if (x) {
1209                         error = (x->km.state == XFRM_STATE_ERROR ?
1210                                  -EINVAL : -EAGAIN);
1211                         xfrm_state_put(x);
1212                 }
1213                 else if (error == -ESRCH)
1214                         error = -EAGAIN;
1215
1216                 if (!tmpl->optional)
1217                         goto fail;
1218         }
1219         return nx;
1220
1221 fail:
1222         for (nx--; nx>=0; nx--)
1223                 xfrm_state_put(xfrm[nx]);
1224         return error;
1225 }
1226
1227 static int
1228 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1229                   struct xfrm_state **xfrm, unsigned short family)
1230 {
1231         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1232         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1233         int cnx = 0;
1234         int error;
1235         int ret;
1236         int i;
1237
1238         for (i = 0; i < npols; i++) {
1239                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1240                         error = -ENOBUFS;
1241                         goto fail;
1242                 }
1243
1244                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1245                 if (ret < 0) {
1246                         error = ret;
1247                         goto fail;
1248                 } else
1249                         cnx += ret;
1250         }
1251
1252         /* found states are sorted for outbound processing */
1253         if (npols > 1)
1254                 xfrm_state_sort(xfrm, tpp, cnx, family);
1255
1256         return cnx;
1257
1258  fail:
1259         for (cnx--; cnx>=0; cnx--)
1260                 xfrm_state_put(tpp[cnx]);
1261         return error;
1262
1263 }
1264
1265 /* Check that the bundle accepts the flow and its components are
1266  * still valid.
1267  */
1268
1269 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1270 {
1271         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1272         int tos;
1273
1274         if (!afinfo)
1275                 return -EINVAL;
1276
1277         tos = afinfo->get_tos(fl);
1278
1279         xfrm_policy_put_afinfo(afinfo);
1280
1281         return tos;
1282 }
1283
1284 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1285 {
1286         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1287         struct dst_entry *dst = &xdst->u.dst;
1288
1289         if (xdst->route == NULL) {
1290                 /* Dummy bundle - if it has xfrms we were not
1291                  * able to build bundle as template resolution failed.
1292                  * It means we need to try again resolving. */
1293                 if (xdst->num_xfrms > 0)
1294                         return NULL;
1295         } else {
1296                 /* Real bundle */
1297                 if (stale_bundle(dst))
1298                         return NULL;
1299         }
1300
1301         dst_hold(dst);
1302         return flo;
1303 }
1304
1305 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1306 {
1307         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1308         struct dst_entry *dst = &xdst->u.dst;
1309
1310         if (!xdst->route)
1311                 return 0;
1312         if (stale_bundle(dst))
1313                 return 0;
1314
1315         return 1;
1316 }
1317
1318 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1319 {
1320         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1321         struct dst_entry *dst = &xdst->u.dst;
1322
1323         dst_free(dst);
1324 }
1325
1326 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1327         .get = xfrm_bundle_flo_get,
1328         .check = xfrm_bundle_flo_check,
1329         .delete = xfrm_bundle_flo_delete,
1330 };
1331
1332 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1333 {
1334         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1335         struct dst_ops *dst_ops;
1336         struct xfrm_dst *xdst;
1337
1338         if (!afinfo)
1339                 return ERR_PTR(-EINVAL);
1340
1341         switch (family) {
1342         case AF_INET:
1343                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1344                 break;
1345 #if IS_ENABLED(CONFIG_IPV6)
1346         case AF_INET6:
1347                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1348                 break;
1349 #endif
1350         default:
1351                 BUG();
1352         }
1353         xdst = dst_alloc(dst_ops, NULL, 0, 0, 0);
1354
1355         if (likely(xdst)) {
1356                 memset(&xdst->u.rt6.rt6i_table, 0,
1357                         sizeof(*xdst) - sizeof(struct dst_entry));
1358                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1359         } else
1360                 xdst = ERR_PTR(-ENOBUFS);
1361
1362         xfrm_policy_put_afinfo(afinfo);
1363
1364         return xdst;
1365 }
1366
1367 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1368                                  int nfheader_len)
1369 {
1370         struct xfrm_policy_afinfo *afinfo =
1371                 xfrm_policy_get_afinfo(dst->ops->family);
1372         int err;
1373
1374         if (!afinfo)
1375                 return -EINVAL;
1376
1377         err = afinfo->init_path(path, dst, nfheader_len);
1378
1379         xfrm_policy_put_afinfo(afinfo);
1380
1381         return err;
1382 }
1383
1384 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1385                                 const struct flowi *fl)
1386 {
1387         struct xfrm_policy_afinfo *afinfo =
1388                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1389         int err;
1390
1391         if (!afinfo)
1392                 return -EINVAL;
1393
1394         err = afinfo->fill_dst(xdst, dev, fl);
1395
1396         xfrm_policy_put_afinfo(afinfo);
1397
1398         return err;
1399 }
1400
1401
1402 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1403  * all the metrics... Shortly, bundle a bundle.
1404  */
1405
1406 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1407                                             struct xfrm_state **xfrm, int nx,
1408                                             const struct flowi *fl,
1409                                             struct dst_entry *dst)
1410 {
1411         struct net *net = xp_net(policy);
1412         unsigned long now = jiffies;
1413         struct net_device *dev;
1414         struct xfrm_mode *inner_mode;
1415         struct dst_entry *dst_prev = NULL;
1416         struct dst_entry *dst0 = NULL;
1417         int i = 0;
1418         int err;
1419         int header_len = 0;
1420         int nfheader_len = 0;
1421         int trailer_len = 0;
1422         int tos;
1423         int family = policy->selector.family;
1424         xfrm_address_t saddr, daddr;
1425
1426         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1427
1428         tos = xfrm_get_tos(fl, family);
1429         err = tos;
1430         if (tos < 0)
1431                 goto put_states;
1432
1433         dst_hold(dst);
1434
1435         for (; i < nx; i++) {
1436                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1437                 struct dst_entry *dst1 = &xdst->u.dst;
1438
1439                 err = PTR_ERR(xdst);
1440                 if (IS_ERR(xdst)) {
1441                         dst_release(dst);
1442                         goto put_states;
1443                 }
1444
1445                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1446                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1447                                                         xfrm_af2proto(family));
1448                         if (!inner_mode) {
1449                                 err = -EAFNOSUPPORT;
1450                                 dst_release(dst);
1451                                 goto put_states;
1452                         }
1453                 } else
1454                         inner_mode = xfrm[i]->inner_mode;
1455
1456                 if (!dst_prev)
1457                         dst0 = dst1;
1458                 else {
1459                         dst_prev->child = dst_clone(dst1);
1460                         dst1->flags |= DST_NOHASH;
1461                 }
1462
1463                 xdst->route = dst;
1464                 dst_copy_metrics(dst1, dst);
1465
1466                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1467                         family = xfrm[i]->props.family;
1468                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1469                                               family);
1470                         err = PTR_ERR(dst);
1471                         if (IS_ERR(dst))
1472                                 goto put_states;
1473                 } else
1474                         dst_hold(dst);
1475
1476                 dst1->xfrm = xfrm[i];
1477                 xdst->xfrm_genid = xfrm[i]->genid;
1478
1479                 dst1->obsolete = -1;
1480                 dst1->flags |= DST_HOST;
1481                 dst1->lastuse = now;
1482
1483                 dst1->input = dst_discard;
1484                 dst1->output = inner_mode->afinfo->output;
1485
1486                 dst1->next = dst_prev;
1487                 dst_prev = dst1;
1488
1489                 header_len += xfrm[i]->props.header_len;
1490                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1491                         nfheader_len += xfrm[i]->props.header_len;
1492                 trailer_len += xfrm[i]->props.trailer_len;
1493         }
1494
1495         dst_prev->child = dst;
1496         dst0->path = dst;
1497
1498         err = -ENODEV;
1499         dev = dst->dev;
1500         if (!dev)
1501                 goto free_dst;
1502
1503         /* Copy neighbour for reachability confirmation */
1504         dst_set_neighbour(dst0, neigh_clone(dst_get_neighbour_noref(dst)));
1505
1506         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1507         xfrm_init_pmtu(dst_prev);
1508
1509         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1510                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1511
1512                 err = xfrm_fill_dst(xdst, dev, fl);
1513                 if (err)
1514                         goto free_dst;
1515
1516                 dst_prev->header_len = header_len;
1517                 dst_prev->trailer_len = trailer_len;
1518                 header_len -= xdst->u.dst.xfrm->props.header_len;
1519                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1520         }
1521
1522 out:
1523         return dst0;
1524
1525 put_states:
1526         for (; i < nx; i++)
1527                 xfrm_state_put(xfrm[i]);
1528 free_dst:
1529         if (dst0)
1530                 dst_free(dst0);
1531         dst0 = ERR_PTR(err);
1532         goto out;
1533 }
1534
1535 static int inline
1536 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1537 {
1538         if (!*target) {
1539                 *target = kmalloc(size, GFP_ATOMIC);
1540                 if (!*target)
1541                         return -ENOMEM;
1542         }
1543         memcpy(*target, src, size);
1544         return 0;
1545 }
1546
1547 static int inline
1548 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1549 {
1550 #ifdef CONFIG_XFRM_SUB_POLICY
1551         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1552         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1553                                    sel, sizeof(*sel));
1554 #else
1555         return 0;
1556 #endif
1557 }
1558
1559 static int inline
1560 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1561 {
1562 #ifdef CONFIG_XFRM_SUB_POLICY
1563         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1564         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1565 #else
1566         return 0;
1567 #endif
1568 }
1569
1570 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1571                                 struct xfrm_policy **pols,
1572                                 int *num_pols, int *num_xfrms)
1573 {
1574         int i;
1575
1576         if (*num_pols == 0 || !pols[0]) {
1577                 *num_pols = 0;
1578                 *num_xfrms = 0;
1579                 return 0;
1580         }
1581         if (IS_ERR(pols[0]))
1582                 return PTR_ERR(pols[0]);
1583
1584         *num_xfrms = pols[0]->xfrm_nr;
1585
1586 #ifdef CONFIG_XFRM_SUB_POLICY
1587         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1588             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1589                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1590                                                     XFRM_POLICY_TYPE_MAIN,
1591                                                     fl, family,
1592                                                     XFRM_POLICY_OUT);
1593                 if (pols[1]) {
1594                         if (IS_ERR(pols[1])) {
1595                                 xfrm_pols_put(pols, *num_pols);
1596                                 return PTR_ERR(pols[1]);
1597                         }
1598                         (*num_pols) ++;
1599                         (*num_xfrms) += pols[1]->xfrm_nr;
1600                 }
1601         }
1602 #endif
1603         for (i = 0; i < *num_pols; i++) {
1604                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1605                         *num_xfrms = -1;
1606                         break;
1607                 }
1608         }
1609
1610         return 0;
1611
1612 }
1613
1614 static struct xfrm_dst *
1615 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1616                                const struct flowi *fl, u16 family,
1617                                struct dst_entry *dst_orig)
1618 {
1619         struct net *net = xp_net(pols[0]);
1620         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1621         struct dst_entry *dst;
1622         struct xfrm_dst *xdst;
1623         int err;
1624
1625         /* Try to instantiate a bundle */
1626         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1627         if (err <= 0) {
1628                 if (err != 0 && err != -EAGAIN)
1629                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1630                 return ERR_PTR(err);
1631         }
1632
1633         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1634         if (IS_ERR(dst)) {
1635                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1636                 return ERR_CAST(dst);
1637         }
1638
1639         xdst = (struct xfrm_dst *)dst;
1640         xdst->num_xfrms = err;
1641         if (num_pols > 1)
1642                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1643         else
1644                 err = xfrm_dst_update_origin(dst, fl);
1645         if (unlikely(err)) {
1646                 dst_free(dst);
1647                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1648                 return ERR_PTR(err);
1649         }
1650
1651         xdst->num_pols = num_pols;
1652         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1653         xdst->policy_genid = atomic_read(&pols[0]->genid);
1654
1655         return xdst;
1656 }
1657
1658 static struct flow_cache_object *
1659 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1660                    struct flow_cache_object *oldflo, void *ctx)
1661 {
1662         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1663         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1664         struct xfrm_dst *xdst, *new_xdst;
1665         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1666
1667         /* Check if the policies from old bundle are usable */
1668         xdst = NULL;
1669         if (oldflo) {
1670                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1671                 num_pols = xdst->num_pols;
1672                 num_xfrms = xdst->num_xfrms;
1673                 pol_dead = 0;
1674                 for (i = 0; i < num_pols; i++) {
1675                         pols[i] = xdst->pols[i];
1676                         pol_dead |= pols[i]->walk.dead;
1677                 }
1678                 if (pol_dead) {
1679                         dst_free(&xdst->u.dst);
1680                         xdst = NULL;
1681                         num_pols = 0;
1682                         num_xfrms = 0;
1683                         oldflo = NULL;
1684                 }
1685         }
1686
1687         /* Resolve policies to use if we couldn't get them from
1688          * previous cache entry */
1689         if (xdst == NULL) {
1690                 num_pols = 1;
1691                 pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1692                 err = xfrm_expand_policies(fl, family, pols,
1693                                            &num_pols, &num_xfrms);
1694                 if (err < 0)
1695                         goto inc_error;
1696                 if (num_pols == 0)
1697                         return NULL;
1698                 if (num_xfrms <= 0)
1699                         goto make_dummy_bundle;
1700         }
1701
1702         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1703         if (IS_ERR(new_xdst)) {
1704                 err = PTR_ERR(new_xdst);
1705                 if (err != -EAGAIN)
1706                         goto error;
1707                 if (oldflo == NULL)
1708                         goto make_dummy_bundle;
1709                 dst_hold(&xdst->u.dst);
1710                 return oldflo;
1711         } else if (new_xdst == NULL) {
1712                 num_xfrms = 0;
1713                 if (oldflo == NULL)
1714                         goto make_dummy_bundle;
1715                 xdst->num_xfrms = 0;
1716                 dst_hold(&xdst->u.dst);
1717                 return oldflo;
1718         }
1719
1720         /* Kill the previous bundle */
1721         if (xdst) {
1722                 /* The policies were stolen for newly generated bundle */
1723                 xdst->num_pols = 0;
1724                 dst_free(&xdst->u.dst);
1725         }
1726
1727         /* Flow cache does not have reference, it dst_free()'s,
1728          * but we do need to return one reference for original caller */
1729         dst_hold(&new_xdst->u.dst);
1730         return &new_xdst->flo;
1731
1732 make_dummy_bundle:
1733         /* We found policies, but there's no bundles to instantiate:
1734          * either because the policy blocks, has no transformations or
1735          * we could not build template (no xfrm_states).*/
1736         xdst = xfrm_alloc_dst(net, family);
1737         if (IS_ERR(xdst)) {
1738                 xfrm_pols_put(pols, num_pols);
1739                 return ERR_CAST(xdst);
1740         }
1741         xdst->num_pols = num_pols;
1742         xdst->num_xfrms = num_xfrms;
1743         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1744
1745         dst_hold(&xdst->u.dst);
1746         return &xdst->flo;
1747
1748 inc_error:
1749         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1750 error:
1751         if (xdst != NULL)
1752                 dst_free(&xdst->u.dst);
1753         else
1754                 xfrm_pols_put(pols, num_pols);
1755         return ERR_PTR(err);
1756 }
1757
1758 static struct dst_entry *make_blackhole(struct net *net, u16 family,
1759                                         struct dst_entry *dst_orig)
1760 {
1761         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1762         struct dst_entry *ret;
1763
1764         if (!afinfo) {
1765                 dst_release(dst_orig);
1766                 ret = ERR_PTR(-EINVAL);
1767         } else {
1768                 ret = afinfo->blackhole_route(net, dst_orig);
1769         }
1770         xfrm_policy_put_afinfo(afinfo);
1771
1772         return ret;
1773 }
1774
1775 /* Main function: finds/creates a bundle for given flow.
1776  *
1777  * At the moment we eat a raw IP route. Mostly to speed up lookups
1778  * on interfaces with disabled IPsec.
1779  */
1780 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
1781                               const struct flowi *fl,
1782                               struct sock *sk, int flags)
1783 {
1784         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1785         struct flow_cache_object *flo;
1786         struct xfrm_dst *xdst;
1787         struct dst_entry *dst, *route;
1788         u16 family = dst_orig->ops->family;
1789         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1790         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
1791
1792 restart:
1793         dst = NULL;
1794         xdst = NULL;
1795         route = NULL;
1796
1797         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1798                 num_pols = 1;
1799                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1800                 err = xfrm_expand_policies(fl, family, pols,
1801                                            &num_pols, &num_xfrms);
1802                 if (err < 0)
1803                         goto dropdst;
1804
1805                 if (num_pols) {
1806                         if (num_xfrms <= 0) {
1807                                 drop_pols = num_pols;
1808                                 goto no_transform;
1809                         }
1810
1811                         xdst = xfrm_resolve_and_create_bundle(
1812                                         pols, num_pols, fl,
1813                                         family, dst_orig);
1814                         if (IS_ERR(xdst)) {
1815                                 xfrm_pols_put(pols, num_pols);
1816                                 err = PTR_ERR(xdst);
1817                                 goto dropdst;
1818                         } else if (xdst == NULL) {
1819                                 num_xfrms = 0;
1820                                 drop_pols = num_pols;
1821                                 goto no_transform;
1822                         }
1823
1824                         dst_hold(&xdst->u.dst);
1825
1826                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
1827                         xdst->u.dst.next = xfrm_policy_sk_bundles;
1828                         xfrm_policy_sk_bundles = &xdst->u.dst;
1829                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
1830
1831                         route = xdst->route;
1832                 }
1833         }
1834
1835         if (xdst == NULL) {
1836                 /* To accelerate a bit...  */
1837                 if ((dst_orig->flags & DST_NOXFRM) ||
1838                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1839                         goto nopol;
1840
1841                 flo = flow_cache_lookup(net, fl, family, dir,
1842                                         xfrm_bundle_lookup, dst_orig);
1843                 if (flo == NULL)
1844                         goto nopol;
1845                 if (IS_ERR(flo)) {
1846                         err = PTR_ERR(flo);
1847                         goto dropdst;
1848                 }
1849                 xdst = container_of(flo, struct xfrm_dst, flo);
1850
1851                 num_pols = xdst->num_pols;
1852                 num_xfrms = xdst->num_xfrms;
1853                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
1854                 route = xdst->route;
1855         }
1856
1857         dst = &xdst->u.dst;
1858         if (route == NULL && num_xfrms > 0) {
1859                 /* The only case when xfrm_bundle_lookup() returns a
1860                  * bundle with null route, is when the template could
1861                  * not be resolved. It means policies are there, but
1862                  * bundle could not be created, since we don't yet
1863                  * have the xfrm_state's. We need to wait for KM to
1864                  * negotiate new SA's or bail out with error.*/
1865                 if (net->xfrm.sysctl_larval_drop) {
1866                         /* EREMOTE tells the caller to generate
1867                          * a one-shot blackhole route. */
1868                         dst_release(dst);
1869                         xfrm_pols_put(pols, drop_pols);
1870                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1871
1872                         return make_blackhole(net, family, dst_orig);
1873                 }
1874                 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
1875                         DECLARE_WAITQUEUE(wait, current);
1876
1877                         add_wait_queue(&net->xfrm.km_waitq, &wait);
1878                         set_current_state(TASK_INTERRUPTIBLE);
1879                         schedule();
1880                         set_current_state(TASK_RUNNING);
1881                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
1882
1883                         if (!signal_pending(current)) {
1884                                 dst_release(dst);
1885                                 goto restart;
1886                         }
1887
1888                         err = -ERESTART;
1889                 } else
1890                         err = -EAGAIN;
1891
1892                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1893                 goto error;
1894         }
1895
1896 no_transform:
1897         if (num_pols == 0)
1898                 goto nopol;
1899
1900         if ((flags & XFRM_LOOKUP_ICMP) &&
1901             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
1902                 err = -ENOENT;
1903                 goto error;
1904         }
1905
1906         for (i = 0; i < num_pols; i++)
1907                 pols[i]->curlft.use_time = get_seconds();
1908
1909         if (num_xfrms < 0) {
1910                 /* Prohibit the flow */
1911                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1912                 err = -EPERM;
1913                 goto error;
1914         } else if (num_xfrms > 0) {
1915                 /* Flow transformed */
1916                 dst_release(dst_orig);
1917         } else {
1918                 /* Flow passes untransformed */
1919                 dst_release(dst);
1920                 dst = dst_orig;
1921         }
1922 ok:
1923         xfrm_pols_put(pols, drop_pols);
1924         return dst;
1925
1926 nopol:
1927         if (!(flags & XFRM_LOOKUP_ICMP)) {
1928                 dst = dst_orig;
1929                 goto ok;
1930         }
1931         err = -ENOENT;
1932 error:
1933         dst_release(dst);
1934 dropdst:
1935         dst_release(dst_orig);
1936         xfrm_pols_put(pols, drop_pols);
1937         return ERR_PTR(err);
1938 }
1939 EXPORT_SYMBOL(xfrm_lookup);
1940
1941 static inline int
1942 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
1943 {
1944         struct xfrm_state *x;
1945
1946         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1947                 return 0;
1948         x = skb->sp->xvec[idx];
1949         if (!x->type->reject)
1950                 return 0;
1951         return x->type->reject(x, skb, fl);
1952 }
1953
1954 /* When skb is transformed back to its "native" form, we have to
1955  * check policy restrictions. At the moment we make this in maximally
1956  * stupid way. Shame on me. :-) Of course, connected sockets must
1957  * have policy cached at them.
1958  */
1959
1960 static inline int
1961 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
1962               unsigned short family)
1963 {
1964         if (xfrm_state_kern(x))
1965                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1966         return  x->id.proto == tmpl->id.proto &&
1967                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1968                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1969                 x->props.mode == tmpl->mode &&
1970                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1971                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1972                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1973                   xfrm_state_addr_cmp(tmpl, x, family));
1974 }
1975
1976 /*
1977  * 0 or more than 0 is returned when validation is succeeded (either bypass
1978  * because of optional transport mode, or next index of the mathced secpath
1979  * state with the template.
1980  * -1 is returned when no matching template is found.
1981  * Otherwise "-2 - errored_index" is returned.
1982  */
1983 static inline int
1984 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
1985                unsigned short family)
1986 {
1987         int idx = start;
1988
1989         if (tmpl->optional) {
1990                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1991                         return start;
1992         } else
1993                 start = -1;
1994         for (; idx < sp->len; idx++) {
1995                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1996                         return ++idx;
1997                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1998                         if (start == -1)
1999                                 start = -2-idx;
2000                         break;
2001                 }
2002         }
2003         return start;
2004 }
2005
2006 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2007                           unsigned int family, int reverse)
2008 {
2009         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2010         int err;
2011
2012         if (unlikely(afinfo == NULL))
2013                 return -EAFNOSUPPORT;
2014
2015         afinfo->decode_session(skb, fl, reverse);
2016         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2017         xfrm_policy_put_afinfo(afinfo);
2018         return err;
2019 }
2020 EXPORT_SYMBOL(__xfrm_decode_session);
2021
2022 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2023 {
2024         for (; k < sp->len; k++) {
2025                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2026                         *idxp = k;
2027                         return 1;
2028                 }
2029         }
2030
2031         return 0;
2032 }
2033
2034 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2035                         unsigned short family)
2036 {
2037         struct net *net = dev_net(skb->dev);
2038         struct xfrm_policy *pol;
2039         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2040         int npols = 0;
2041         int xfrm_nr;
2042         int pi;
2043         int reverse;
2044         struct flowi fl;
2045         u8 fl_dir;
2046         int xerr_idx = -1;
2047
2048         reverse = dir & ~XFRM_POLICY_MASK;
2049         dir &= XFRM_POLICY_MASK;
2050         fl_dir = policy_to_flow_dir(dir);
2051
2052         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2053                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2054                 return 0;
2055         }
2056
2057         nf_nat_decode_session(skb, &fl, family);
2058
2059         /* First, check used SA against their selectors. */
2060         if (skb->sp) {
2061                 int i;
2062
2063                 for (i=skb->sp->len-1; i>=0; i--) {
2064                         struct xfrm_state *x = skb->sp->xvec[i];
2065                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2066                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2067                                 return 0;
2068                         }
2069                 }
2070         }
2071
2072         pol = NULL;
2073         if (sk && sk->sk_policy[dir]) {
2074                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2075                 if (IS_ERR(pol)) {
2076                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2077                         return 0;
2078                 }
2079         }
2080
2081         if (!pol) {
2082                 struct flow_cache_object *flo;
2083
2084                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2085                                         xfrm_policy_lookup, NULL);
2086                 if (IS_ERR_OR_NULL(flo))
2087                         pol = ERR_CAST(flo);
2088                 else
2089                         pol = container_of(flo, struct xfrm_policy, flo);
2090         }
2091
2092         if (IS_ERR(pol)) {
2093                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2094                 return 0;
2095         }
2096
2097         if (!pol) {
2098                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2099                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2100                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2101                         return 0;
2102                 }
2103                 return 1;
2104         }
2105
2106         pol->curlft.use_time = get_seconds();
2107
2108         pols[0] = pol;
2109         npols ++;
2110 #ifdef CONFIG_XFRM_SUB_POLICY
2111         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2112                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2113                                                     &fl, family,
2114                                                     XFRM_POLICY_IN);
2115                 if (pols[1]) {
2116                         if (IS_ERR(pols[1])) {
2117                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2118                                 return 0;
2119                         }
2120                         pols[1]->curlft.use_time = get_seconds();
2121                         npols ++;
2122                 }
2123         }
2124 #endif
2125
2126         if (pol->action == XFRM_POLICY_ALLOW) {
2127                 struct sec_path *sp;
2128                 static struct sec_path dummy;
2129                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2130                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2131                 struct xfrm_tmpl **tpp = tp;
2132                 int ti = 0;
2133                 int i, k;
2134
2135                 if ((sp = skb->sp) == NULL)
2136                         sp = &dummy;
2137
2138                 for (pi = 0; pi < npols; pi++) {
2139                         if (pols[pi] != pol &&
2140                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2141                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2142                                 goto reject;
2143                         }
2144                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2145                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2146                                 goto reject_error;
2147                         }
2148                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2149                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2150                 }
2151                 xfrm_nr = ti;
2152                 if (npols > 1) {
2153                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2154                         tpp = stp;
2155                 }
2156
2157                 /* For each tunnel xfrm, find the first matching tmpl.
2158                  * For each tmpl before that, find corresponding xfrm.
2159                  * Order is _important_. Later we will implement
2160                  * some barriers, but at the moment barriers
2161                  * are implied between each two transformations.
2162                  */
2163                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2164                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2165                         if (k < 0) {
2166                                 if (k < -1)
2167                                         /* "-2 - errored_index" returned */
2168                                         xerr_idx = -(2+k);
2169                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2170                                 goto reject;
2171                         }
2172                 }
2173
2174                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2175                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2176                         goto reject;
2177                 }
2178
2179                 xfrm_pols_put(pols, npols);
2180                 return 1;
2181         }
2182         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2183
2184 reject:
2185         xfrm_secpath_reject(xerr_idx, skb, &fl);
2186 reject_error:
2187         xfrm_pols_put(pols, npols);
2188         return 0;
2189 }
2190 EXPORT_SYMBOL(__xfrm_policy_check);
2191
2192 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2193 {
2194         struct net *net = dev_net(skb->dev);
2195         struct flowi fl;
2196         struct dst_entry *dst;
2197         int res = 1;
2198
2199         if (xfrm_decode_session(skb, &fl, family) < 0) {
2200                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2201                 return 0;
2202         }
2203
2204         skb_dst_force(skb);
2205
2206         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2207         if (IS_ERR(dst)) {
2208                 res = 0;
2209                 dst = NULL;
2210         }
2211         skb_dst_set(skb, dst);
2212         return res;
2213 }
2214 EXPORT_SYMBOL(__xfrm_route_forward);
2215
2216 /* Optimize later using cookies and generation ids. */
2217
2218 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2219 {
2220         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2221          * to "-1" to force all XFRM destinations to get validated by
2222          * dst_ops->check on every use.  We do this because when a
2223          * normal route referenced by an XFRM dst is obsoleted we do
2224          * not go looking around for all parent referencing XFRM dsts
2225          * so that we can invalidate them.  It is just too much work.
2226          * Instead we make the checks here on every use.  For example:
2227          *
2228          *      XFRM dst A --> IPv4 dst X
2229          *
2230          * X is the "xdst->route" of A (X is also the "dst->path" of A
2231          * in this example).  If X is marked obsolete, "A" will not
2232          * notice.  That's what we are validating here via the
2233          * stale_bundle() check.
2234          *
2235          * When a policy's bundle is pruned, we dst_free() the XFRM
2236          * dst which causes it's ->obsolete field to be set to a
2237          * positive non-zero integer.  If an XFRM dst has been pruned
2238          * like this, we want to force a new route lookup.
2239          */
2240         if (dst->obsolete < 0 && !stale_bundle(dst))
2241                 return dst;
2242
2243         return NULL;
2244 }
2245
2246 static int stale_bundle(struct dst_entry *dst)
2247 {
2248         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2249 }
2250
2251 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2252 {
2253         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2254                 dst->dev = dev_net(dev)->loopback_dev;
2255                 dev_hold(dst->dev);
2256                 dev_put(dev);
2257         }
2258 }
2259 EXPORT_SYMBOL(xfrm_dst_ifdown);
2260
2261 static void xfrm_link_failure(struct sk_buff *skb)
2262 {
2263         /* Impossible. Such dst must be popped before reaches point of failure. */
2264 }
2265
2266 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2267 {
2268         if (dst) {
2269                 if (dst->obsolete) {
2270                         dst_release(dst);
2271                         dst = NULL;
2272                 }
2273         }
2274         return dst;
2275 }
2276
2277 static void __xfrm_garbage_collect(struct net *net)
2278 {
2279         struct dst_entry *head, *next;
2280
2281         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2282         head = xfrm_policy_sk_bundles;
2283         xfrm_policy_sk_bundles = NULL;
2284         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2285
2286         while (head) {
2287                 next = head->next;
2288                 dst_free(head);
2289                 head = next;
2290         }
2291 }
2292
2293 static void xfrm_garbage_collect(struct net *net)
2294 {
2295         flow_cache_flush();
2296         __xfrm_garbage_collect(net);
2297 }
2298
2299 static void xfrm_garbage_collect_deferred(struct net *net)
2300 {
2301         flow_cache_flush_deferred();
2302         __xfrm_garbage_collect(net);
2303 }
2304
2305 static void xfrm_init_pmtu(struct dst_entry *dst)
2306 {
2307         do {
2308                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2309                 u32 pmtu, route_mtu_cached;
2310
2311                 pmtu = dst_mtu(dst->child);
2312                 xdst->child_mtu_cached = pmtu;
2313
2314                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2315
2316                 route_mtu_cached = dst_mtu(xdst->route);
2317                 xdst->route_mtu_cached = route_mtu_cached;
2318
2319                 if (pmtu > route_mtu_cached)
2320                         pmtu = route_mtu_cached;
2321
2322                 dst_metric_set(dst, RTAX_MTU, pmtu);
2323         } while ((dst = dst->next));
2324 }
2325
2326 /* Check that the bundle accepts the flow and its components are
2327  * still valid.
2328  */
2329
2330 static int xfrm_bundle_ok(struct xfrm_dst *first)
2331 {
2332         struct dst_entry *dst = &first->u.dst;
2333         struct xfrm_dst *last;
2334         u32 mtu;
2335
2336         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2337             (dst->dev && !netif_running(dst->dev)))
2338                 return 0;
2339
2340         last = NULL;
2341
2342         do {
2343                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2344
2345                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2346                         return 0;
2347                 if (xdst->xfrm_genid != dst->xfrm->genid)
2348                         return 0;
2349                 if (xdst->num_pols > 0 &&
2350                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2351                         return 0;
2352
2353                 mtu = dst_mtu(dst->child);
2354                 if (xdst->child_mtu_cached != mtu) {
2355                         last = xdst;
2356                         xdst->child_mtu_cached = mtu;
2357                 }
2358
2359                 if (!dst_check(xdst->route, xdst->route_cookie))
2360                         return 0;
2361                 mtu = dst_mtu(xdst->route);
2362                 if (xdst->route_mtu_cached != mtu) {
2363                         last = xdst;
2364                         xdst->route_mtu_cached = mtu;
2365                 }
2366
2367                 dst = dst->child;
2368         } while (dst->xfrm);
2369
2370         if (likely(!last))
2371                 return 1;
2372
2373         mtu = last->child_mtu_cached;
2374         for (;;) {
2375                 dst = &last->u.dst;
2376
2377                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2378                 if (mtu > last->route_mtu_cached)
2379                         mtu = last->route_mtu_cached;
2380                 dst_metric_set(dst, RTAX_MTU, mtu);
2381
2382                 if (last == first)
2383                         break;
2384
2385                 last = (struct xfrm_dst *)last->u.dst.next;
2386                 last->child_mtu_cached = mtu;
2387         }
2388
2389         return 1;
2390 }
2391
2392 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2393 {
2394         return dst_metric_advmss(dst->path);
2395 }
2396
2397 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2398 {
2399         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2400
2401         return mtu ? : dst_mtu(dst->path);
2402 }
2403
2404 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst, const void *daddr)
2405 {
2406         return dst_neigh_lookup(dst->path, daddr);
2407 }
2408
2409 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2410 {
2411         struct net *net;
2412         int err = 0;
2413         if (unlikely(afinfo == NULL))
2414                 return -EINVAL;
2415         if (unlikely(afinfo->family >= NPROTO))
2416                 return -EAFNOSUPPORT;
2417         write_lock_bh(&xfrm_policy_afinfo_lock);
2418         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2419                 err = -ENOBUFS;
2420         else {
2421                 struct dst_ops *dst_ops = afinfo->dst_ops;
2422                 if (likely(dst_ops->kmem_cachep == NULL))
2423                         dst_ops->kmem_cachep = xfrm_dst_cache;
2424                 if (likely(dst_ops->check == NULL))
2425                         dst_ops->check = xfrm_dst_check;
2426                 if (likely(dst_ops->default_advmss == NULL))
2427                         dst_ops->default_advmss = xfrm_default_advmss;
2428                 if (likely(dst_ops->mtu == NULL))
2429                         dst_ops->mtu = xfrm_mtu;
2430                 if (likely(dst_ops->negative_advice == NULL))
2431                         dst_ops->negative_advice = xfrm_negative_advice;
2432                 if (likely(dst_ops->link_failure == NULL))
2433                         dst_ops->link_failure = xfrm_link_failure;
2434                 if (likely(dst_ops->neigh_lookup == NULL))
2435                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2436                 if (likely(afinfo->garbage_collect == NULL))
2437                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2438                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2439         }
2440         write_unlock_bh(&xfrm_policy_afinfo_lock);
2441
2442         rtnl_lock();
2443         for_each_net(net) {
2444                 struct dst_ops *xfrm_dst_ops;
2445
2446                 switch (afinfo->family) {
2447                 case AF_INET:
2448                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2449                         break;
2450 #if IS_ENABLED(CONFIG_IPV6)
2451                 case AF_INET6:
2452                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2453                         break;
2454 #endif
2455                 default:
2456                         BUG();
2457                 }
2458                 *xfrm_dst_ops = *afinfo->dst_ops;
2459         }
2460         rtnl_unlock();
2461
2462         return err;
2463 }
2464 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2465
2466 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2467 {
2468         int err = 0;
2469         if (unlikely(afinfo == NULL))
2470                 return -EINVAL;
2471         if (unlikely(afinfo->family >= NPROTO))
2472                 return -EAFNOSUPPORT;
2473         write_lock_bh(&xfrm_policy_afinfo_lock);
2474         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2475                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2476                         err = -EINVAL;
2477                 else {
2478                         struct dst_ops *dst_ops = afinfo->dst_ops;
2479                         xfrm_policy_afinfo[afinfo->family] = NULL;
2480                         dst_ops->kmem_cachep = NULL;
2481                         dst_ops->check = NULL;
2482                         dst_ops->negative_advice = NULL;
2483                         dst_ops->link_failure = NULL;
2484                         afinfo->garbage_collect = NULL;
2485                 }
2486         }
2487         write_unlock_bh(&xfrm_policy_afinfo_lock);
2488         return err;
2489 }
2490 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2491
2492 static void __net_init xfrm_dst_ops_init(struct net *net)
2493 {
2494         struct xfrm_policy_afinfo *afinfo;
2495
2496         read_lock_bh(&xfrm_policy_afinfo_lock);
2497         afinfo = xfrm_policy_afinfo[AF_INET];
2498         if (afinfo)
2499                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2500 #if IS_ENABLED(CONFIG_IPV6)
2501         afinfo = xfrm_policy_afinfo[AF_INET6];
2502         if (afinfo)
2503                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2504 #endif
2505         read_unlock_bh(&xfrm_policy_afinfo_lock);
2506 }
2507
2508 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2509 {
2510         struct xfrm_policy_afinfo *afinfo;
2511         if (unlikely(family >= NPROTO))
2512                 return NULL;
2513         read_lock(&xfrm_policy_afinfo_lock);
2514         afinfo = xfrm_policy_afinfo[family];
2515         if (unlikely(!afinfo))
2516                 read_unlock(&xfrm_policy_afinfo_lock);
2517         return afinfo;
2518 }
2519
2520 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2521 {
2522         read_unlock(&xfrm_policy_afinfo_lock);
2523 }
2524
2525 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2526 {
2527         struct net_device *dev = ptr;
2528
2529         switch (event) {
2530         case NETDEV_DOWN:
2531                 xfrm_garbage_collect(dev_net(dev));
2532         }
2533         return NOTIFY_DONE;
2534 }
2535
2536 static struct notifier_block xfrm_dev_notifier = {
2537         .notifier_call  = xfrm_dev_event,
2538 };
2539
2540 #ifdef CONFIG_XFRM_STATISTICS
2541 static int __net_init xfrm_statistics_init(struct net *net)
2542 {
2543         int rv;
2544
2545         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2546                           sizeof(struct linux_xfrm_mib),
2547                           __alignof__(struct linux_xfrm_mib)) < 0)
2548                 return -ENOMEM;
2549         rv = xfrm_proc_init(net);
2550         if (rv < 0)
2551                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2552         return rv;
2553 }
2554
2555 static void xfrm_statistics_fini(struct net *net)
2556 {
2557         xfrm_proc_fini(net);
2558         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2559 }
2560 #else
2561 static int __net_init xfrm_statistics_init(struct net *net)
2562 {
2563         return 0;
2564 }
2565
2566 static void xfrm_statistics_fini(struct net *net)
2567 {
2568 }
2569 #endif
2570
2571 static int __net_init xfrm_policy_init(struct net *net)
2572 {
2573         unsigned int hmask, sz;
2574         int dir;
2575
2576         if (net_eq(net, &init_net))
2577                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2578                                            sizeof(struct xfrm_dst),
2579                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2580                                            NULL);
2581
2582         hmask = 8 - 1;
2583         sz = (hmask+1) * sizeof(struct hlist_head);
2584
2585         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2586         if (!net->xfrm.policy_byidx)
2587                 goto out_byidx;
2588         net->xfrm.policy_idx_hmask = hmask;
2589
2590         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2591                 struct xfrm_policy_hash *htab;
2592
2593                 net->xfrm.policy_count[dir] = 0;
2594                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2595
2596                 htab = &net->xfrm.policy_bydst[dir];
2597                 htab->table = xfrm_hash_alloc(sz);
2598                 if (!htab->table)
2599                         goto out_bydst;
2600                 htab->hmask = hmask;
2601         }
2602
2603         INIT_LIST_HEAD(&net->xfrm.policy_all);
2604         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2605         if (net_eq(net, &init_net))
2606                 register_netdevice_notifier(&xfrm_dev_notifier);
2607         return 0;
2608
2609 out_bydst:
2610         for (dir--; dir >= 0; dir--) {
2611                 struct xfrm_policy_hash *htab;
2612
2613                 htab = &net->xfrm.policy_bydst[dir];
2614                 xfrm_hash_free(htab->table, sz);
2615         }
2616         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2617 out_byidx:
2618         return -ENOMEM;
2619 }
2620
2621 static void xfrm_policy_fini(struct net *net)
2622 {
2623         struct xfrm_audit audit_info;
2624         unsigned int sz;
2625         int dir;
2626
2627         flush_work(&net->xfrm.policy_hash_work);
2628 #ifdef CONFIG_XFRM_SUB_POLICY
2629         audit_info.loginuid = -1;
2630         audit_info.sessionid = -1;
2631         audit_info.secid = 0;
2632         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2633 #endif
2634         audit_info.loginuid = -1;
2635         audit_info.sessionid = -1;
2636         audit_info.secid = 0;
2637         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2638
2639         WARN_ON(!list_empty(&net->xfrm.policy_all));
2640
2641         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2642                 struct xfrm_policy_hash *htab;
2643
2644                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2645
2646                 htab = &net->xfrm.policy_bydst[dir];
2647                 sz = (htab->hmask + 1);
2648                 WARN_ON(!hlist_empty(htab->table));
2649                 xfrm_hash_free(htab->table, sz);
2650         }
2651
2652         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2653         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2654         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2655 }
2656
2657 static int __net_init xfrm_net_init(struct net *net)
2658 {
2659         int rv;
2660
2661         rv = xfrm_statistics_init(net);
2662         if (rv < 0)
2663                 goto out_statistics;
2664         rv = xfrm_state_init(net);
2665         if (rv < 0)
2666                 goto out_state;
2667         rv = xfrm_policy_init(net);
2668         if (rv < 0)
2669                 goto out_policy;
2670         xfrm_dst_ops_init(net);
2671         rv = xfrm_sysctl_init(net);
2672         if (rv < 0)
2673                 goto out_sysctl;
2674         return 0;
2675
2676 out_sysctl:
2677         xfrm_policy_fini(net);
2678 out_policy:
2679         xfrm_state_fini(net);
2680 out_state:
2681         xfrm_statistics_fini(net);
2682 out_statistics:
2683         return rv;
2684 }
2685
2686 static void __net_exit xfrm_net_exit(struct net *net)
2687 {
2688         xfrm_sysctl_fini(net);
2689         xfrm_policy_fini(net);
2690         xfrm_state_fini(net);
2691         xfrm_statistics_fini(net);
2692 }
2693
2694 static struct pernet_operations __net_initdata xfrm_net_ops = {
2695         .init = xfrm_net_init,
2696         .exit = xfrm_net_exit,
2697 };
2698
2699 void __init xfrm_init(void)
2700 {
2701         register_pernet_subsys(&xfrm_net_ops);
2702         xfrm_input_init();
2703 }
2704
2705 #ifdef CONFIG_AUDITSYSCALL
2706 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2707                                          struct audit_buffer *audit_buf)
2708 {
2709         struct xfrm_sec_ctx *ctx = xp->security;
2710         struct xfrm_selector *sel = &xp->selector;
2711
2712         if (ctx)
2713                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2714                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2715
2716         switch(sel->family) {
2717         case AF_INET:
2718                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2719                 if (sel->prefixlen_s != 32)
2720                         audit_log_format(audit_buf, " src_prefixlen=%d",
2721                                          sel->prefixlen_s);
2722                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2723                 if (sel->prefixlen_d != 32)
2724                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2725                                          sel->prefixlen_d);
2726                 break;
2727         case AF_INET6:
2728                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2729                 if (sel->prefixlen_s != 128)
2730                         audit_log_format(audit_buf, " src_prefixlen=%d",
2731                                          sel->prefixlen_s);
2732                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2733                 if (sel->prefixlen_d != 128)
2734                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2735                                          sel->prefixlen_d);
2736                 break;
2737         }
2738 }
2739
2740 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2741                            uid_t auid, u32 sessionid, u32 secid)
2742 {
2743         struct audit_buffer *audit_buf;
2744
2745         audit_buf = xfrm_audit_start("SPD-add");
2746         if (audit_buf == NULL)
2747                 return;
2748         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2749         audit_log_format(audit_buf, " res=%u", result);
2750         xfrm_audit_common_policyinfo(xp, audit_buf);
2751         audit_log_end(audit_buf);
2752 }
2753 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2754
2755 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2756                               uid_t auid, u32 sessionid, u32 secid)
2757 {
2758         struct audit_buffer *audit_buf;
2759
2760         audit_buf = xfrm_audit_start("SPD-delete");
2761         if (audit_buf == NULL)
2762                 return;
2763         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2764         audit_log_format(audit_buf, " res=%u", result);
2765         xfrm_audit_common_policyinfo(xp, audit_buf);
2766         audit_log_end(audit_buf);
2767 }
2768 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2769 #endif
2770
2771 #ifdef CONFIG_XFRM_MIGRATE
2772 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
2773                                         const struct xfrm_selector *sel_tgt)
2774 {
2775         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2776                 if (sel_tgt->family == sel_cmp->family &&
2777                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2778                                   sel_cmp->family) == 0 &&
2779                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2780                                   sel_cmp->family) == 0 &&
2781                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2782                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2783                         return true;
2784                 }
2785         } else {
2786                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2787                         return true;
2788                 }
2789         }
2790         return false;
2791 }
2792
2793 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
2794                                                      u8 dir, u8 type)
2795 {
2796         struct xfrm_policy *pol, *ret = NULL;
2797         struct hlist_node *entry;
2798         struct hlist_head *chain;
2799         u32 priority = ~0U;
2800
2801         read_lock_bh(&xfrm_policy_lock);
2802         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2803         hlist_for_each_entry(pol, entry, chain, bydst) {
2804                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2805                     pol->type == type) {
2806                         ret = pol;
2807                         priority = ret->priority;
2808                         break;
2809                 }
2810         }
2811         chain = &init_net.xfrm.policy_inexact[dir];
2812         hlist_for_each_entry(pol, entry, chain, bydst) {
2813                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2814                     pol->type == type &&
2815                     pol->priority < priority) {
2816                         ret = pol;
2817                         break;
2818                 }
2819         }
2820
2821         if (ret)
2822                 xfrm_pol_hold(ret);
2823
2824         read_unlock_bh(&xfrm_policy_lock);
2825
2826         return ret;
2827 }
2828
2829 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
2830 {
2831         int match = 0;
2832
2833         if (t->mode == m->mode && t->id.proto == m->proto &&
2834             (m->reqid == 0 || t->reqid == m->reqid)) {
2835                 switch (t->mode) {
2836                 case XFRM_MODE_TUNNEL:
2837                 case XFRM_MODE_BEET:
2838                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2839                                           m->old_family) == 0 &&
2840                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2841                                           m->old_family) == 0) {
2842                                 match = 1;
2843                         }
2844                         break;
2845                 case XFRM_MODE_TRANSPORT:
2846                         /* in case of transport mode, template does not store
2847                            any IP addresses, hence we just compare mode and
2848                            protocol */
2849                         match = 1;
2850                         break;
2851                 default:
2852                         break;
2853                 }
2854         }
2855         return match;
2856 }
2857
2858 /* update endpoint address(es) of template(s) */
2859 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2860                                struct xfrm_migrate *m, int num_migrate)
2861 {
2862         struct xfrm_migrate *mp;
2863         int i, j, n = 0;
2864
2865         write_lock_bh(&pol->lock);
2866         if (unlikely(pol->walk.dead)) {
2867                 /* target policy has been deleted */
2868                 write_unlock_bh(&pol->lock);
2869                 return -ENOENT;
2870         }
2871
2872         for (i = 0; i < pol->xfrm_nr; i++) {
2873                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2874                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2875                                 continue;
2876                         n++;
2877                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2878                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2879                                 continue;
2880                         /* update endpoints */
2881                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2882                                sizeof(pol->xfrm_vec[i].id.daddr));
2883                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2884                                sizeof(pol->xfrm_vec[i].saddr));
2885                         pol->xfrm_vec[i].encap_family = mp->new_family;
2886                         /* flush bundles */
2887                         atomic_inc(&pol->genid);
2888                 }
2889         }
2890
2891         write_unlock_bh(&pol->lock);
2892
2893         if (!n)
2894                 return -ENODATA;
2895
2896         return 0;
2897 }
2898
2899 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
2900 {
2901         int i, j;
2902
2903         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2904                 return -EINVAL;
2905
2906         for (i = 0; i < num_migrate; i++) {
2907                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2908                                    m[i].old_family) == 0) &&
2909                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2910                                    m[i].old_family) == 0))
2911                         return -EINVAL;
2912                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2913                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2914                         return -EINVAL;
2915
2916                 /* check if there is any duplicated entry */
2917                 for (j = i + 1; j < num_migrate; j++) {
2918                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2919                                     sizeof(m[i].old_daddr)) &&
2920                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2921                                     sizeof(m[i].old_saddr)) &&
2922                             m[i].proto == m[j].proto &&
2923                             m[i].mode == m[j].mode &&
2924                             m[i].reqid == m[j].reqid &&
2925                             m[i].old_family == m[j].old_family)
2926                                 return -EINVAL;
2927                 }
2928         }
2929
2930         return 0;
2931 }
2932
2933 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
2934                  struct xfrm_migrate *m, int num_migrate,
2935                  struct xfrm_kmaddress *k)
2936 {
2937         int i, err, nx_cur = 0, nx_new = 0;
2938         struct xfrm_policy *pol = NULL;
2939         struct xfrm_state *x, *xc;
2940         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2941         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2942         struct xfrm_migrate *mp;
2943
2944         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2945                 goto out;
2946
2947         /* Stage 1 - find policy */
2948         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2949                 err = -ENOENT;
2950                 goto out;
2951         }
2952
2953         /* Stage 2 - find and update state(s) */
2954         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2955                 if ((x = xfrm_migrate_state_find(mp))) {
2956                         x_cur[nx_cur] = x;
2957                         nx_cur++;
2958                         if ((xc = xfrm_state_migrate(x, mp))) {
2959                                 x_new[nx_new] = xc;
2960                                 nx_new++;
2961                         } else {
2962                                 err = -ENODATA;
2963                                 goto restore_state;
2964                         }
2965                 }
2966         }
2967
2968         /* Stage 3 - update policy */
2969         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2970                 goto restore_state;
2971
2972         /* Stage 4 - delete old state(s) */
2973         if (nx_cur) {
2974                 xfrm_states_put(x_cur, nx_cur);
2975                 xfrm_states_delete(x_cur, nx_cur);
2976         }
2977
2978         /* Stage 5 - announce */
2979         km_migrate(sel, dir, type, m, num_migrate, k);
2980
2981         xfrm_pol_put(pol);
2982
2983         return 0;
2984 out:
2985         return err;
2986
2987 restore_state:
2988         if (pol)
2989                 xfrm_pol_put(pol);
2990         if (nx_cur)
2991                 xfrm_states_put(x_cur, nx_cur);
2992         if (nx_new)
2993                 xfrm_states_delete(x_new, nx_new);
2994
2995         return err;
2996 }
2997 EXPORT_SYMBOL(xfrm_migrate);
2998 #endif