Pileus Git - ~andy/linux/blob - 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 int
  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 int
  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 int 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 0;
  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 match, ret = -ESRCH;
 882
 883         if (pol->family != family ||
 884             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
 885             pol->type != type)
 886                 return ret;
 887
 888         match = xfrm_selector_match(sel, fl, family);
 889         if (match)
 890                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
 891                                                   dir);
 892
 893         return ret;
 894 }
 895
 896 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 897                                                      const struct flowi *fl,
 898                                                      u16 family, u8 dir)
 899 {
 900         int err;
 901         struct xfrm_policy *pol, *ret;
 902         const xfrm_address_t *daddr, *saddr;
 903         struct hlist_node *entry;
 904         struct hlist_head *chain;
 905         u32 priority = ~0U;
 906
 907         daddr = xfrm_flowi_daddr(fl, family);
 908         saddr = xfrm_flowi_saddr(fl, family);
 909         if (unlikely(!daddr || !saddr))
 910                 return NULL;
 911
 912         read_lock_bh(&xfrm_policy_lock);
 913         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 914         ret = NULL;
 915         hlist_for_each_entry(pol, entry, chain, bydst) {
 916                 err = xfrm_policy_match(pol, fl, type, family, dir);
 917                 if (err) {
 918                         if (err == -ESRCH)
 919                                 continue;
 920                         else {
 921                                 ret = ERR_PTR(err);
 922                                 goto fail;
 923                         }
 924                 } else {
 925                         ret = pol;
 926                         priority = ret->priority;
 927                         break;
 928                 }
 929         }
 930         chain = &net->xfrm.policy_inexact[dir];
 931         hlist_for_each_entry(pol, entry, chain, bydst) {
 932                 err = xfrm_policy_match(pol, fl, type, family, dir);
 933                 if (err) {
 934                         if (err == -ESRCH)
 935                                 continue;
 936                         else {
 937                                 ret = ERR_PTR(err);
 938                                 goto fail;
 939                         }
 940                 } else if (pol->priority < priority) {
 941                         ret = pol;
 942                         break;
 943                 }
 944         }
 945         if (ret)
 946                 xfrm_pol_hold(ret);
 947 fail:
 948         read_unlock_bh(&xfrm_policy_lock);
 949
 950         return ret;
 951 }
 952
 953 static struct xfrm_policy *
 954 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
 955 {
 956 #ifdef CONFIG_XFRM_SUB_POLICY
 957         struct xfrm_policy *pol;
 958
 959         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
 960         if (pol != NULL)
 961                 return pol;
 962 #endif
 963         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
 964 }
 965
 966 static struct flow_cache_object *
 967 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
 968                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
 969 {
 970         struct xfrm_policy *pol;
 971
 972         if (old_obj)
 973                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
 974
 975         pol = __xfrm_policy_lookup(net, fl, family, dir);
 976         if (IS_ERR_OR_NULL(pol))
 977                 return ERR_CAST(pol);
 978
 979         /* Resolver returns two references:
 980          * one for cache and one for caller of flow_cache_lookup() */
 981         xfrm_pol_hold(pol);
 982
 983         return &pol->flo;
 984 }
 985
 986 static inline int policy_to_flow_dir(int dir)
 987 {
 988         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
 989             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
 990             XFRM_POLICY_FWD == FLOW_DIR_FWD)
 991                 return dir;
 992         switch (dir) {
 993         default:
 994         case XFRM_POLICY_IN:
 995                 return FLOW_DIR_IN;
 996         case XFRM_POLICY_OUT:
 997                 return FLOW_DIR_OUT;
 998         case XFRM_POLICY_FWD:
 999                 return FLOW_DIR_FWD;
1000         }
1001 }
1002
1003 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1004                                                  const struct flowi *fl)
1005 {
1006         struct xfrm_policy *pol;
1007
1008         read_lock_bh(&xfrm_policy_lock);
1009         if ((pol = sk->sk_policy[dir]) != NULL) {
1010                 int match = xfrm_selector_match(&pol->selector, fl,
1011                                                 sk->sk_family);
1012                 int err = 0;
1013
1014                 if (match) {
1015                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1016                                 pol = NULL;
1017                                 goto out;
1018                         }
1019                         err = security_xfrm_policy_lookup(pol->security,
1020                                                       fl->flowi_secid,
1021                                                       policy_to_flow_dir(dir));
1022                         if (!err)
1023                                 xfrm_pol_hold(pol);
1024                         else if (err == -ESRCH)
1025                                 pol = NULL;
1026                         else
1027                                 pol = ERR_PTR(err);
1028                 } else
1029                         pol = NULL;
1030         }
1031 out:
1032         read_unlock_bh(&xfrm_policy_lock);
1033         return pol;
1034 }
1035
1036 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1037 {
1038         struct net *net = xp_net(pol);
1039         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1040                                                      pol->family, dir);
1041
1042         list_add(&pol->walk.all, &net->xfrm.policy_all);
1043         hlist_add_head(&pol->bydst, chain);
1044         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1045         net->xfrm.policy_count[dir]++;
1046         xfrm_pol_hold(pol);
1047
1048         if (xfrm_bydst_should_resize(net, dir, NULL))
1049                 schedule_work(&net->xfrm.policy_hash_work);
1050 }
1051
1052 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1053                                                 int dir)
1054 {
1055         struct net *net = xp_net(pol);
1056
1057         if (hlist_unhashed(&pol->bydst))
1058                 return NULL;
1059
1060         hlist_del(&pol->bydst);
1061         hlist_del(&pol->byidx);
1062         list_del(&pol->walk.all);
1063         net->xfrm.policy_count[dir]--;
1064
1065         return pol;
1066 }
1067
1068 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1069 {
1070         write_lock_bh(&xfrm_policy_lock);
1071         pol = __xfrm_policy_unlink(pol, dir);
1072         write_unlock_bh(&xfrm_policy_lock);
1073         if (pol) {
1074                 xfrm_policy_kill(pol);
1075                 return 0;
1076         }
1077         return -ENOENT;
1078 }
1079 EXPORT_SYMBOL(xfrm_policy_delete);
1080
1081 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1082 {
1083         struct net *net = xp_net(pol);
1084         struct xfrm_policy *old_pol;
1085
1086 #ifdef CONFIG_XFRM_SUB_POLICY
1087         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1088                 return -EINVAL;
1089 #endif
1090
1091         write_lock_bh(&xfrm_policy_lock);
1092         old_pol = sk->sk_policy[dir];
1093         sk->sk_policy[dir] = pol;
1094         if (pol) {
1095                 pol->curlft.add_time = get_seconds();
1096                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir);
1097                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1098         }
1099         if (old_pol)
1100                 /* Unlinking succeeds always. This is the only function
1101                  * allowed to delete or replace socket policy.
1102                  */
1103                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1104         write_unlock_bh(&xfrm_policy_lock);
1105
1106         if (old_pol) {
1107                 xfrm_policy_kill(old_pol);
1108         }
1109         return 0;
1110 }
1111
1112 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1113 {
1114         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1115
1116         if (newp) {
1117                 newp->selector = old->selector;
1118                 if (security_xfrm_policy_clone(old->security,
1119                                                &newp->security)) {
1120                         kfree(newp);
1121                         return NULL;  /* ENOMEM */
1122                 }
1123                 newp->lft = old->lft;
1124                 newp->curlft = old->curlft;
1125                 newp->mark = old->mark;
1126                 newp->action = old->action;
1127                 newp->flags = old->flags;
1128                 newp->xfrm_nr = old->xfrm_nr;
1129                 newp->index = old->index;
1130                 newp->type = old->type;
1131                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1132                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1133                 write_lock_bh(&xfrm_policy_lock);
1134                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1135                 write_unlock_bh(&xfrm_policy_lock);
1136                 xfrm_pol_put(newp);
1137         }
1138         return newp;
1139 }
1140
1141 int __xfrm_sk_clone_policy(struct sock *sk)
1142 {
1143         struct xfrm_policy *p0 = sk->sk_policy[0],
1144                            *p1 = sk->sk_policy[1];
1145
1146         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1147         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1148                 return -ENOMEM;
1149         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1150                 return -ENOMEM;
1151         return 0;
1152 }
1153
1154 static int
1155 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1156                unsigned short family)
1157 {
1158         int err;
1159         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1160
1161         if (unlikely(afinfo == NULL))
1162                 return -EINVAL;
1163         err = afinfo->get_saddr(net, local, remote);
1164         xfrm_policy_put_afinfo(afinfo);
1165         return err;
1166 }
1167
1168 /* Resolve list of templates for the flow, given policy. */
1169
1170 static int
1171 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1172                       struct xfrm_state **xfrm, unsigned short family)
1173 {
1174         struct net *net = xp_net(policy);
1175         int nx;
1176         int i, error;
1177         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1178         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1179         xfrm_address_t tmp;
1180
1181         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1182                 struct xfrm_state *x;
1183                 xfrm_address_t *remote = daddr;
1184                 xfrm_address_t *local  = saddr;
1185                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1186
1187                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1188                     tmpl->mode == XFRM_MODE_BEET) {
1189                         remote = &tmpl->id.daddr;
1190                         local = &tmpl->saddr;
1191                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1192                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1193                                 if (error)
1194                                         goto fail;
1195                                 local = &tmp;
1196                         }
1197                 }
1198
1199                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1200
1201                 if (x && x->km.state == XFRM_STATE_VALID) {
1202                         xfrm[nx++] = x;
1203                         daddr = remote;
1204                         saddr = local;
1205                         continue;
1206                 }
1207                 if (x) {
1208                         error = (x->km.state == XFRM_STATE_ERROR ?
1209                                  -EINVAL : -EAGAIN);
1210                         xfrm_state_put(x);
1211                 }
1212                 else if (error == -ESRCH)
1213                         error = -EAGAIN;
1214
1215                 if (!tmpl->optional)
1216                         goto fail;
1217         }
1218         return nx;
1219
1220 fail:
1221         for (nx--; nx>=0; nx--)
1222                 xfrm_state_put(xfrm[nx]);
1223         return error;
1224 }
1225
1226 static int
1227 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1228                   struct xfrm_state **xfrm, unsigned short family)
1229 {
1230         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1231         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1232         int cnx = 0;
1233         int error;
1234         int ret;
1235         int i;
1236
1237         for (i = 0; i < npols; i++) {
1238                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1239                         error = -ENOBUFS;
1240                         goto fail;
1241                 }
1242
1243                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1244                 if (ret < 0) {
1245                         error = ret;
1246                         goto fail;
1247                 } else
1248                         cnx += ret;
1249         }
1250
1251         /* found states are sorted for outbound processing */
1252         if (npols > 1)
1253                 xfrm_state_sort(xfrm, tpp, cnx, family);
1254
1255         return cnx;
1256
1257  fail:
1258         for (cnx--; cnx>=0; cnx--)
1259                 xfrm_state_put(tpp[cnx]);
1260         return error;
1261
1262 }
1263
1264 /* Check that the bundle accepts the flow and its components are
1265  * still valid.
1266  */
1267
1268 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1269 {
1270         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1271         int tos;
1272
1273         if (!afinfo)
1274                 return -EINVAL;
1275
1276         tos = afinfo->get_tos(fl);
1277
1278         xfrm_policy_put_afinfo(afinfo);
1279
1280         return tos;
1281 }
1282
1283 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1284 {
1285         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1286         struct dst_entry *dst = &xdst->u.dst;
1287
1288         if (xdst->route == NULL) {
1289                 /* Dummy bundle - if it has xfrms we were not
1290                  * able to build bundle as template resolution failed.
1291                  * It means we need to try again resolving. */
1292                 if (xdst->num_xfrms > 0)
1293                         return NULL;
1294         } else {
1295                 /* Real bundle */
1296                 if (stale_bundle(dst))
1297                         return NULL;
1298         }
1299
1300         dst_hold(dst);
1301         return flo;
1302 }
1303
1304 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1305 {
1306         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1307         struct dst_entry *dst = &xdst->u.dst;
1308
1309         if (!xdst->route)
1310                 return 0;
1311         if (stale_bundle(dst))
1312                 return 0;
1313
1314         return 1;
1315 }
1316
1317 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1318 {
1319         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1320         struct dst_entry *dst = &xdst->u.dst;
1321
1322         dst_free(dst);
1323 }
1324
1325 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1326         .get = xfrm_bundle_flo_get,
1327         .check = xfrm_bundle_flo_check,
1328         .delete = xfrm_bundle_flo_delete,
1329 };
1330
1331 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1332 {
1333         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1334         struct dst_ops *dst_ops;
1335         struct xfrm_dst *xdst;
1336
1337         if (!afinfo)
1338                 return ERR_PTR(-EINVAL);
1339
1340         switch (family) {
1341         case AF_INET:
1342                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1343                 break;
1344 #if IS_ENABLED(CONFIG_IPV6)
1345         case AF_INET6:
1346                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1347                 break;
1348 #endif
1349         default:
1350                 BUG();
1351         }
1352         xdst = dst_alloc(dst_ops, NULL, 0, 0, 0);
1353
1354         if (likely(xdst)) {
1355                 memset(&xdst->u.rt6.rt6i_table, 0,
1356                         sizeof(*xdst) - sizeof(struct dst_entry));
1357                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1358         } else
1359                 xdst = ERR_PTR(-ENOBUFS);
1360
1361         xfrm_policy_put_afinfo(afinfo);
1362
1363         return xdst;
1364 }
1365
1366 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1367                                  int nfheader_len)
1368 {
1369         struct xfrm_policy_afinfo *afinfo =
1370                 xfrm_policy_get_afinfo(dst->ops->family);
1371         int err;
1372
1373         if (!afinfo)
1374                 return -EINVAL;
1375
1376         err = afinfo->init_path(path, dst, nfheader_len);
1377
1378         xfrm_policy_put_afinfo(afinfo);
1379
1380         return err;
1381 }
1382
1383 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1384                                 const struct flowi *fl)
1385 {
1386         struct xfrm_policy_afinfo *afinfo =
1387                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1388         int err;
1389
1390         if (!afinfo)
1391                 return -EINVAL;
1392
1393         err = afinfo->fill_dst(xdst, dev, fl);
1394
1395         xfrm_policy_put_afinfo(afinfo);
1396
1397         return err;
1398 }
1399
1400
1401 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1402  * all the metrics... Shortly, bundle a bundle.
1403  */
1404
1405 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1406                                             struct xfrm_state **xfrm, int nx,
1407                                             const struct flowi *fl,
1408                                             struct dst_entry *dst)
1409 {
1410         struct net *net = xp_net(policy);
1411         unsigned long now = jiffies;
1412         struct net_device *dev;
1413         struct xfrm_mode *inner_mode;
1414         struct dst_entry *dst_prev = NULL;
1415         struct dst_entry *dst0 = NULL;
1416         int i = 0;
1417         int err;
1418         int header_len = 0;
1419         int nfheader_len = 0;
1420         int trailer_len = 0;
1421         int tos;
1422         int family = policy->selector.family;
1423         xfrm_address_t saddr, daddr;
1424
1425         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1426
1427         tos = xfrm_get_tos(fl, family);
1428         err = tos;
1429         if (tos < 0)
1430                 goto put_states;
1431
1432         dst_hold(dst);
1433
1434         for (; i < nx; i++) {
1435                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1436                 struct dst_entry *dst1 = &xdst->u.dst;
1437
1438                 err = PTR_ERR(xdst);
1439                 if (IS_ERR(xdst)) {
1440                         dst_release(dst);
1441                         goto put_states;
1442                 }
1443
1444                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1445                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1446                                                         xfrm_af2proto(family));
1447                         if (!inner_mode) {
1448                                 err = -EAFNOSUPPORT;
1449                                 dst_release(dst);
1450                                 goto put_states;
1451                         }
1452                 } else
1453                         inner_mode = xfrm[i]->inner_mode;
1454
1455                 if (!dst_prev)
1456                         dst0 = dst1;
1457                 else {
1458                         dst_prev->child = dst_clone(dst1);
1459                         dst1->flags |= DST_NOHASH;
1460                 }
1461
1462                 xdst->route = dst;
1463                 dst_copy_metrics(dst1, dst);
1464
1465                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1466                         family = xfrm[i]->props.family;
1467                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1468                                               family);
1469                         err = PTR_ERR(dst);
1470                         if (IS_ERR(dst))
1471                                 goto put_states;
1472                 } else
1473                         dst_hold(dst);
1474
1475                 dst1->xfrm = xfrm[i];
1476                 xdst->xfrm_genid = xfrm[i]->genid;
1477
1478                 dst1->obsolete = -1;
1479                 dst1->flags |= DST_HOST;
1480                 dst1->lastuse = now;
1481
1482                 dst1->input = dst_discard;
1483                 dst1->output = inner_mode->afinfo->output;
1484
1485                 dst1->next = dst_prev;
1486                 dst_prev = dst1;
1487
1488                 header_len += xfrm[i]->props.header_len;
1489                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1490                         nfheader_len += xfrm[i]->props.header_len;
1491                 trailer_len += xfrm[i]->props.trailer_len;
1492         }
1493
1494         dst_prev->child = dst;
1495         dst0->path = dst;
1496
1497         err = -ENODEV;
1498         dev = dst->dev;
1499         if (!dev)
1500                 goto free_dst;
1501
1502         /* Copy neighbour for reachability confirmation */
1503         dst_set_neighbour(dst0, neigh_clone(dst_get_neighbour_noref(dst)));
1504
1505         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1506         xfrm_init_pmtu(dst_prev);
1507
1508         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1509                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1510
1511                 err = xfrm_fill_dst(xdst, dev, fl);
1512                 if (err)
1513                         goto free_dst;
1514
1515                 dst_prev->header_len = header_len;
1516                 dst_prev->trailer_len = trailer_len;
1517                 header_len -= xdst->u.dst.xfrm->props.header_len;
1518                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1519         }
1520
1521 out:
1522         return dst0;
1523
1524 put_states:
1525         for (; i < nx; i++)
1526                 xfrm_state_put(xfrm[i]);
1527 free_dst:
1528         if (dst0)
1529                 dst_free(dst0);
1530         dst0 = ERR_PTR(err);
1531         goto out;
1532 }
1533
1534 static int inline
1535 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1536 {
1537         if (!*target) {
1538                 *target = kmalloc(size, GFP_ATOMIC);
1539                 if (!*target)
1540                         return -ENOMEM;
1541         }
1542         memcpy(*target, src, size);
1543         return 0;
1544 }
1545
1546 static int inline
1547 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1548 {
1549 #ifdef CONFIG_XFRM_SUB_POLICY
1550         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1551         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1552                                    sel, sizeof(*sel));
1553 #else
1554         return 0;
1555 #endif
1556 }
1557
1558 static int inline
1559 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1560 {
1561 #ifdef CONFIG_XFRM_SUB_POLICY
1562         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1563         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1564 #else
1565         return 0;
1566 #endif
1567 }
1568
1569 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1570                                 struct xfrm_policy **pols,
1571                                 int *num_pols, int *num_xfrms)
1572 {
1573         int i;
1574
1575         if (*num_pols == 0 || !pols[0]) {
1576                 *num_pols = 0;
1577                 *num_xfrms = 0;
1578                 return 0;
1579         }
1580         if (IS_ERR(pols[0]))
1581                 return PTR_ERR(pols[0]);
1582
1583         *num_xfrms = pols[0]->xfrm_nr;
1584
1585 #ifdef CONFIG_XFRM_SUB_POLICY
1586         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1587             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1588                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1589                                                     XFRM_POLICY_TYPE_MAIN,
1590                                                     fl, family,
1591                                                     XFRM_POLICY_OUT);
1592                 if (pols[1]) {
1593                         if (IS_ERR(pols[1])) {
1594                                 xfrm_pols_put(pols, *num_pols);
1595                                 return PTR_ERR(pols[1]);
1596                         }
1597                         (*num_pols) ++;
1598                         (*num_xfrms) += pols[1]->xfrm_nr;
1599                 }
1600         }
1601 #endif
1602         for (i = 0; i < *num_pols; i++) {
1603                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1604                         *num_xfrms = -1;
1605                         break;
1606                 }
1607         }
1608
1609         return 0;
1610
1611 }
1612
1613 static struct xfrm_dst *
1614 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1615                                const struct flowi *fl, u16 family,
1616                                struct dst_entry *dst_orig)
1617 {
1618         struct net *net = xp_net(pols[0]);
1619         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1620         struct dst_entry *dst;
1621         struct xfrm_dst *xdst;
1622         int err;
1623
1624         /* Try to instantiate a bundle */
1625         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1626         if (err <= 0) {
1627                 if (err != 0 && err != -EAGAIN)
1628                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1629                 return ERR_PTR(err);
1630         }
1631
1632         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1633         if (IS_ERR(dst)) {
1634                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1635                 return ERR_CAST(dst);
1636         }
1637
1638         xdst = (struct xfrm_dst *)dst;
1639         xdst->num_xfrms = err;
1640         if (num_pols > 1)
1641                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1642         else
1643                 err = xfrm_dst_update_origin(dst, fl);
1644         if (unlikely(err)) {
1645                 dst_free(dst);
1646                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1647                 return ERR_PTR(err);
1648         }
1649
1650         xdst->num_pols = num_pols;
1651         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1652         xdst->policy_genid = atomic_read(&pols[0]->genid);
1653
1654         return xdst;
1655 }
1656
1657 static struct flow_cache_object *
1658 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1659                    struct flow_cache_object *oldflo, void *ctx)
1660 {
1661         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1662         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1663         struct xfrm_dst *xdst, *new_xdst;
1664         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1665
1666         /* Check if the policies from old bundle are usable */
1667         xdst = NULL;
1668         if (oldflo) {
1669                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1670                 num_pols = xdst->num_pols;
1671                 num_xfrms = xdst->num_xfrms;
1672                 pol_dead = 0;
1673                 for (i = 0; i < num_pols; i++) {
1674                         pols[i] = xdst->pols[i];
1675                         pol_dead |= pols[i]->walk.dead;
1676                 }
1677                 if (pol_dead) {
1678                         dst_free(&xdst->u.dst);
1679                         xdst = NULL;
1680                         num_pols = 0;
1681                         num_xfrms = 0;
1682                         oldflo = NULL;
1683                 }
1684         }
1685
1686         /* Resolve policies to use if we couldn't get them from
1687          * previous cache entry */
1688         if (xdst == NULL) {
1689                 num_pols = 1;
1690                 pols[0] = __xfrm_policy_lookup(net, fl, family, dir);
1691                 err = xfrm_expand_policies(fl, family, pols,
1692                                            &num_pols, &num_xfrms);
1693                 if (err < 0)
1694                         goto inc_error;
1695                 if (num_pols == 0)
1696                         return NULL;
1697                 if (num_xfrms <= 0)
1698                         goto make_dummy_bundle;
1699         }
1700
1701         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1702         if (IS_ERR(new_xdst)) {
1703                 err = PTR_ERR(new_xdst);
1704                 if (err != -EAGAIN)
1705                         goto error;
1706                 if (oldflo == NULL)
1707                         goto make_dummy_bundle;
1708                 dst_hold(&xdst->u.dst);
1709                 return oldflo;
1710         } else if (new_xdst == NULL) {
1711                 num_xfrms = 0;
1712                 if (oldflo == NULL)
1713                         goto make_dummy_bundle;
1714                 xdst->num_xfrms = 0;
1715                 dst_hold(&xdst->u.dst);
1716                 return oldflo;
1717         }
1718
1719         /* Kill the previous bundle */
1720         if (xdst) {
1721                 /* The policies were stolen for newly generated bundle */
1722                 xdst->num_pols = 0;
1723                 dst_free(&xdst->u.dst);
1724         }
1725
1726         /* Flow cache does not have reference, it dst_free()'s,
1727          * but we do need to return one reference for original caller */
1728         dst_hold(&new_xdst->u.dst);
1729         return &new_xdst->flo;
1730
1731 make_dummy_bundle:
1732         /* We found policies, but there's no bundles to instantiate:
1733          * either because the policy blocks, has no transformations or
1734          * we could not build template (no xfrm_states).*/
1735         xdst = xfrm_alloc_dst(net, family);
1736         if (IS_ERR(xdst)) {
1737                 xfrm_pols_put(pols, num_pols);
1738                 return ERR_CAST(xdst);
1739         }
1740         xdst->num_pols = num_pols;
1741         xdst->num_xfrms = num_xfrms;
1742         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1743
1744         dst_hold(&xdst->u.dst);
1745         return &xdst->flo;
1746
1747 inc_error:
1748         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1749 error:
1750         if (xdst != NULL)
1751                 dst_free(&xdst->u.dst);
1752         else
1753                 xfrm_pols_put(pols, num_pols);
1754         return ERR_PTR(err);
1755 }
1756
1757 static struct dst_entry *make_blackhole(struct net *net, u16 family,
1758                                         struct dst_entry *dst_orig)
1759 {
1760         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1761         struct dst_entry *ret;
1762
1763         if (!afinfo) {
1764                 dst_release(dst_orig);
1765                 ret = ERR_PTR(-EINVAL);
1766         } else {
1767                 ret = afinfo->blackhole_route(net, dst_orig);
1768         }
1769         xfrm_policy_put_afinfo(afinfo);
1770
1771         return ret;
1772 }
1773
1774 /* Main function: finds/creates a bundle for given flow.
1775  *
1776  * At the moment we eat a raw IP route. Mostly to speed up lookups
1777  * on interfaces with disabled IPsec.
1778  */
1779 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
1780                               const struct flowi *fl,
1781                               struct sock *sk, int flags)
1782 {
1783         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1784         struct flow_cache_object *flo;
1785         struct xfrm_dst *xdst;
1786         struct dst_entry *dst, *route;
1787         u16 family = dst_orig->ops->family;
1788         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1789         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
1790
1791 restart:
1792         dst = NULL;
1793         xdst = NULL;
1794         route = NULL;
1795
1796         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
1797                 num_pols = 1;
1798                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1799                 err = xfrm_expand_policies(fl, family, pols,
1800                                            &num_pols, &num_xfrms);
1801                 if (err < 0)
1802                         goto dropdst;
1803
1804                 if (num_pols) {
1805                         if (num_xfrms <= 0) {
1806                                 drop_pols = num_pols;
1807                                 goto no_transform;
1808                         }
1809
1810                         xdst = xfrm_resolve_and_create_bundle(
1811                                         pols, num_pols, fl,
1812                                         family, dst_orig);
1813                         if (IS_ERR(xdst)) {
1814                                 xfrm_pols_put(pols, num_pols);
1815                                 err = PTR_ERR(xdst);
1816                                 goto dropdst;
1817                         } else if (xdst == NULL) {
1818                                 num_xfrms = 0;
1819                                 drop_pols = num_pols;
1820                                 goto no_transform;
1821                         }
1822
1823                         dst_hold(&xdst->u.dst);
1824
1825                         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
1826                         xdst->u.dst.next = xfrm_policy_sk_bundles;
1827                         xfrm_policy_sk_bundles = &xdst->u.dst;
1828                         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
1829
1830                         route = xdst->route;
1831                 }
1832         }
1833
1834         if (xdst == NULL) {
1835                 /* To accelerate a bit...  */
1836                 if ((dst_orig->flags & DST_NOXFRM) ||
1837                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
1838                         goto nopol;
1839
1840                 flo = flow_cache_lookup(net, fl, family, dir,
1841                                         xfrm_bundle_lookup, dst_orig);
1842                 if (flo == NULL)
1843                         goto nopol;
1844                 if (IS_ERR(flo)) {
1845                         err = PTR_ERR(flo);
1846                         goto dropdst;
1847                 }
1848                 xdst = container_of(flo, struct xfrm_dst, flo);
1849
1850                 num_pols = xdst->num_pols;
1851                 num_xfrms = xdst->num_xfrms;
1852                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
1853                 route = xdst->route;
1854         }
1855
1856         dst = &xdst->u.dst;
1857         if (route == NULL && num_xfrms > 0) {
1858                 /* The only case when xfrm_bundle_lookup() returns a
1859                  * bundle with null route, is when the template could
1860                  * not be resolved. It means policies are there, but
1861                  * bundle could not be created, since we don't yet
1862                  * have the xfrm_state's. We need to wait for KM to
1863                  * negotiate new SA's or bail out with error.*/
1864                 if (net->xfrm.sysctl_larval_drop) {
1865                         /* EREMOTE tells the caller to generate
1866                          * a one-shot blackhole route. */
1867                         dst_release(dst);
1868                         xfrm_pols_put(pols, drop_pols);
1869                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1870
1871                         return make_blackhole(net, family, dst_orig);
1872                 }
1873                 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
1874                         DECLARE_WAITQUEUE(wait, current);
1875
1876                         add_wait_queue(&net->xfrm.km_waitq, &wait);
1877                         set_current_state(TASK_INTERRUPTIBLE);
1878                         schedule();
1879                         set_current_state(TASK_RUNNING);
1880                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
1881
1882                         if (!signal_pending(current)) {
1883                                 dst_release(dst);
1884                                 goto restart;
1885                         }
1886
1887                         err = -ERESTART;
1888                 } else
1889                         err = -EAGAIN;
1890
1891                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
1892                 goto error;
1893         }
1894
1895 no_transform:
1896         if (num_pols == 0)
1897                 goto nopol;
1898
1899         if ((flags & XFRM_LOOKUP_ICMP) &&
1900             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
1901                 err = -ENOENT;
1902                 goto error;
1903         }
1904
1905         for (i = 0; i < num_pols; i++)
1906                 pols[i]->curlft.use_time = get_seconds();
1907
1908         if (num_xfrms < 0) {
1909                 /* Prohibit the flow */
1910                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
1911                 err = -EPERM;
1912                 goto error;
1913         } else if (num_xfrms > 0) {
1914                 /* Flow transformed */
1915                 dst_release(dst_orig);
1916         } else {
1917                 /* Flow passes untransformed */
1918                 dst_release(dst);
1919                 dst = dst_orig;
1920         }
1921 ok:
1922         xfrm_pols_put(pols, drop_pols);
1923         return dst;
1924
1925 nopol:
1926         if (!(flags & XFRM_LOOKUP_ICMP)) {
1927                 dst = dst_orig;
1928                 goto ok;
1929         }
1930         err = -ENOENT;
1931 error:
1932         dst_release(dst);
1933 dropdst:
1934         dst_release(dst_orig);
1935         xfrm_pols_put(pols, drop_pols);
1936         return ERR_PTR(err);
1937 }
1938 EXPORT_SYMBOL(xfrm_lookup);
1939
1940 static inline int
1941 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
1942 {
1943         struct xfrm_state *x;
1944
1945         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1946                 return 0;
1947         x = skb->sp->xvec[idx];
1948         if (!x->type->reject)
1949                 return 0;
1950         return x->type->reject(x, skb, fl);
1951 }
1952
1953 /* When skb is transformed back to its "native" form, we have to
1954  * check policy restrictions. At the moment we make this in maximally
1955  * stupid way. Shame on me. :-) Of course, connected sockets must
1956  * have policy cached at them.
1957  */
1958
1959 static inline int
1960 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
1961               unsigned short family)
1962 {
1963         if (xfrm_state_kern(x))
1964                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
1965         return  x->id.proto == tmpl->id.proto &&
1966                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1967                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1968                 x->props.mode == tmpl->mode &&
1969                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
1970                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1971                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1972                   xfrm_state_addr_cmp(tmpl, x, family));
1973 }
1974
1975 /*
1976  * 0 or more than 0 is returned when validation is succeeded (either bypass
1977  * because of optional transport mode, or next index of the mathced secpath
1978  * state with the template.
1979  * -1 is returned when no matching template is found.
1980  * Otherwise "-2 - errored_index" is returned.
1981  */
1982 static inline int
1983 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
1984                unsigned short family)
1985 {
1986         int idx = start;
1987
1988         if (tmpl->optional) {
1989                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1990                         return start;
1991         } else
1992                 start = -1;
1993         for (; idx < sp->len; idx++) {
1994                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1995                         return ++idx;
1996                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1997                         if (start == -1)
1998                                 start = -2-idx;
1999                         break;
2000                 }
2001         }
2002         return start;
2003 }
2004
2005 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2006                           unsigned int family, int reverse)
2007 {
2008         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2009         int err;
2010
2011         if (unlikely(afinfo == NULL))
2012                 return -EAFNOSUPPORT;
2013
2014         afinfo->decode_session(skb, fl, reverse);
2015         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2016         xfrm_policy_put_afinfo(afinfo);
2017         return err;
2018 }
2019 EXPORT_SYMBOL(__xfrm_decode_session);
2020
2021 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2022 {
2023         for (; k < sp->len; k++) {
2024                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2025                         *idxp = k;
2026                         return 1;
2027                 }
2028         }
2029
2030         return 0;
2031 }
2032
2033 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2034                         unsigned short family)
2035 {
2036         struct net *net = dev_net(skb->dev);
2037         struct xfrm_policy *pol;
2038         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2039         int npols = 0;
2040         int xfrm_nr;
2041         int pi;
2042         int reverse;
2043         struct flowi fl;
2044         u8 fl_dir;
2045         int xerr_idx = -1;
2046
2047         reverse = dir & ~XFRM_POLICY_MASK;
2048         dir &= XFRM_POLICY_MASK;
2049         fl_dir = policy_to_flow_dir(dir);
2050
2051         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2052                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2053                 return 0;
2054         }
2055
2056         nf_nat_decode_session(skb, &fl, family);
2057
2058         /* First, check used SA against their selectors. */
2059         if (skb->sp) {
2060                 int i;
2061
2062                 for (i=skb->sp->len-1; i>=0; i--) {
2063                         struct xfrm_state *x = skb->sp->xvec[i];
2064                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2065                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2066                                 return 0;
2067                         }
2068                 }
2069         }
2070
2071         pol = NULL;
2072         if (sk && sk->sk_policy[dir]) {
2073                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2074                 if (IS_ERR(pol)) {
2075                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2076                         return 0;
2077                 }
2078         }
2079
2080         if (!pol) {
2081                 struct flow_cache_object *flo;
2082
2083                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2084                                         xfrm_policy_lookup, NULL);
2085                 if (IS_ERR_OR_NULL(flo))
2086                         pol = ERR_CAST(flo);
2087                 else
2088                         pol = container_of(flo, struct xfrm_policy, flo);
2089         }
2090
2091         if (IS_ERR(pol)) {
2092                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2093                 return 0;
2094         }
2095
2096         if (!pol) {
2097                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2098                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2099                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2100                         return 0;
2101                 }
2102                 return 1;
2103         }
2104
2105         pol->curlft.use_time = get_seconds();
2106
2107         pols[0] = pol;
2108         npols ++;
2109 #ifdef CONFIG_XFRM_SUB_POLICY
2110         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2111                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2112                                                     &fl, family,
2113                                                     XFRM_POLICY_IN);
2114                 if (pols[1]) {
2115                         if (IS_ERR(pols[1])) {
2116                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2117                                 return 0;
2118                         }
2119                         pols[1]->curlft.use_time = get_seconds();
2120                         npols ++;
2121                 }
2122         }
2123 #endif
2124
2125         if (pol->action == XFRM_POLICY_ALLOW) {
2126                 struct sec_path *sp;
2127                 static struct sec_path dummy;
2128                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2129                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2130                 struct xfrm_tmpl **tpp = tp;
2131                 int ti = 0;
2132                 int i, k;
2133
2134                 if ((sp = skb->sp) == NULL)
2135                         sp = &dummy;
2136
2137                 for (pi = 0; pi < npols; pi++) {
2138                         if (pols[pi] != pol &&
2139                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2140                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2141                                 goto reject;
2142                         }
2143                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2144                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2145                                 goto reject_error;
2146                         }
2147                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2148                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2149                 }
2150                 xfrm_nr = ti;
2151                 if (npols > 1) {
2152                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
2153                         tpp = stp;
2154                 }
2155
2156                 /* For each tunnel xfrm, find the first matching tmpl.
2157                  * For each tmpl before that, find corresponding xfrm.
2158                  * Order is _important_. Later we will implement
2159                  * some barriers, but at the moment barriers
2160                  * are implied between each two transformations.
2161                  */
2162                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2163                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2164                         if (k < 0) {
2165                                 if (k < -1)
2166                                         /* "-2 - errored_index" returned */
2167                                         xerr_idx = -(2+k);
2168                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2169                                 goto reject;
2170                         }
2171                 }
2172
2173                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2174                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2175                         goto reject;
2176                 }
2177
2178                 xfrm_pols_put(pols, npols);
2179                 return 1;
2180         }
2181         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2182
2183 reject:
2184         xfrm_secpath_reject(xerr_idx, skb, &fl);
2185 reject_error:
2186         xfrm_pols_put(pols, npols);
2187         return 0;
2188 }
2189 EXPORT_SYMBOL(__xfrm_policy_check);
2190
2191 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2192 {
2193         struct net *net = dev_net(skb->dev);
2194         struct flowi fl;
2195         struct dst_entry *dst;
2196         int res = 1;
2197
2198         if (xfrm_decode_session(skb, &fl, family) < 0) {
2199                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2200                 return 0;
2201         }
2202
2203         skb_dst_force(skb);
2204
2205         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2206         if (IS_ERR(dst)) {
2207                 res = 0;
2208                 dst = NULL;
2209         }
2210         skb_dst_set(skb, dst);
2211         return res;
2212 }
2213 EXPORT_SYMBOL(__xfrm_route_forward);
2214
2215 /* Optimize later using cookies and generation ids. */
2216
2217 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2218 {
2219         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2220          * to "-1" to force all XFRM destinations to get validated by
2221          * dst_ops->check on every use.  We do this because when a
2222          * normal route referenced by an XFRM dst is obsoleted we do
2223          * not go looking around for all parent referencing XFRM dsts
2224          * so that we can invalidate them.  It is just too much work.
2225          * Instead we make the checks here on every use.  For example:
2226          *
2227          *      XFRM dst A --> IPv4 dst X
2228          *
2229          * X is the "xdst->route" of A (X is also the "dst->path" of A
2230          * in this example).  If X is marked obsolete, "A" will not
2231          * notice.  That's what we are validating here via the
2232          * stale_bundle() check.
2233          *
2234          * When a policy's bundle is pruned, we dst_free() the XFRM
2235          * dst which causes it's ->obsolete field to be set to a
2236          * positive non-zero integer.  If an XFRM dst has been pruned
2237          * like this, we want to force a new route lookup.
2238          */
2239         if (dst->obsolete < 0 && !stale_bundle(dst))
2240                 return dst;
2241
2242         return NULL;
2243 }
2244
2245 static int stale_bundle(struct dst_entry *dst)
2246 {
2247         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2248 }
2249
2250 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2251 {
2252         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2253                 dst->dev = dev_net(dev)->loopback_dev;
2254                 dev_hold(dst->dev);
2255                 dev_put(dev);
2256         }
2257 }
2258 EXPORT_SYMBOL(xfrm_dst_ifdown);
2259
2260 static void xfrm_link_failure(struct sk_buff *skb)
2261 {
2262         /* Impossible. Such dst must be popped before reaches point of failure. */
2263 }
2264
2265 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2266 {
2267         if (dst) {
2268                 if (dst->obsolete) {
2269                         dst_release(dst);
2270                         dst = NULL;
2271                 }
2272         }
2273         return dst;
2274 }
2275
2276 static void __xfrm_garbage_collect(struct net *net)
2277 {
2278         struct dst_entry *head, *next;
2279
2280         spin_lock_bh(&xfrm_policy_sk_bundle_lock);
2281         head = xfrm_policy_sk_bundles;
2282         xfrm_policy_sk_bundles = NULL;
2283         spin_unlock_bh(&xfrm_policy_sk_bundle_lock);
2284
2285         while (head) {
2286                 next = head->next;
2287                 dst_free(head);
2288                 head = next;
2289         }
2290 }
2291
2292 static void xfrm_garbage_collect(struct net *net)
2293 {
2294         flow_cache_flush();
2295         __xfrm_garbage_collect(net);
2296 }
2297
2298 static void xfrm_garbage_collect_deferred(struct net *net)
2299 {
2300         flow_cache_flush_deferred();
2301         __xfrm_garbage_collect(net);
2302 }
2303
2304 static void xfrm_init_pmtu(struct dst_entry *dst)
2305 {
2306         do {
2307                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2308                 u32 pmtu, route_mtu_cached;
2309
2310                 pmtu = dst_mtu(dst->child);
2311                 xdst->child_mtu_cached = pmtu;
2312
2313                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2314
2315                 route_mtu_cached = dst_mtu(xdst->route);
2316                 xdst->route_mtu_cached = route_mtu_cached;
2317
2318                 if (pmtu > route_mtu_cached)
2319                         pmtu = route_mtu_cached;
2320
2321                 dst_metric_set(dst, RTAX_MTU, pmtu);
2322         } while ((dst = dst->next));
2323 }
2324
2325 /* Check that the bundle accepts the flow and its components are
2326  * still valid.
2327  */
2328
2329 static int xfrm_bundle_ok(struct xfrm_dst *first)
2330 {
2331         struct dst_entry *dst = &first->u.dst;
2332         struct xfrm_dst *last;
2333         u32 mtu;
2334
2335         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2336             (dst->dev && !netif_running(dst->dev)))
2337                 return 0;
2338
2339         last = NULL;
2340
2341         do {
2342                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2343
2344                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2345                         return 0;
2346                 if (xdst->xfrm_genid != dst->xfrm->genid)
2347                         return 0;
2348                 if (xdst->num_pols > 0 &&
2349                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2350                         return 0;
2351
2352                 mtu = dst_mtu(dst->child);
2353                 if (xdst->child_mtu_cached != mtu) {
2354                         last = xdst;
2355                         xdst->child_mtu_cached = mtu;
2356                 }
2357
2358                 if (!dst_check(xdst->route, xdst->route_cookie))
2359                         return 0;
2360                 mtu = dst_mtu(xdst->route);
2361                 if (xdst->route_mtu_cached != mtu) {
2362                         last = xdst;
2363                         xdst->route_mtu_cached = mtu;
2364                 }
2365
2366                 dst = dst->child;
2367         } while (dst->xfrm);
2368
2369         if (likely(!last))
2370                 return 1;
2371
2372         mtu = last->child_mtu_cached;
2373         for (;;) {
2374                 dst = &last->u.dst;
2375
2376                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2377                 if (mtu > last->route_mtu_cached)
2378                         mtu = last->route_mtu_cached;
2379                 dst_metric_set(dst, RTAX_MTU, mtu);
2380
2381                 if (last == first)
2382                         break;
2383
2384                 last = (struct xfrm_dst *)last->u.dst.next;
2385                 last->child_mtu_cached = mtu;
2386         }
2387
2388         return 1;
2389 }
2390
2391 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2392 {
2393         return dst_metric_advmss(dst->path);
2394 }
2395
2396 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2397 {
2398         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2399
2400         return mtu ? : dst_mtu(dst->path);
2401 }
2402
2403 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst, const void *daddr)
2404 {
2405         return dst_neigh_lookup(dst->path, daddr);
2406 }
2407
2408 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2409 {
2410         struct net *net;
2411         int err = 0;
2412         if (unlikely(afinfo == NULL))
2413                 return -EINVAL;
2414         if (unlikely(afinfo->family >= NPROTO))
2415                 return -EAFNOSUPPORT;
2416         write_lock_bh(&xfrm_policy_afinfo_lock);
2417         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2418                 err = -ENOBUFS;
2419         else {
2420                 struct dst_ops *dst_ops = afinfo->dst_ops;
2421                 if (likely(dst_ops->kmem_cachep == NULL))
2422                         dst_ops->kmem_cachep = xfrm_dst_cache;
2423                 if (likely(dst_ops->check == NULL))
2424                         dst_ops->check = xfrm_dst_check;
2425                 if (likely(dst_ops->default_advmss == NULL))
2426                         dst_ops->default_advmss = xfrm_default_advmss;
2427                 if (likely(dst_ops->mtu == NULL))
2428                         dst_ops->mtu = xfrm_mtu;
2429                 if (likely(dst_ops->negative_advice == NULL))
2430                         dst_ops->negative_advice = xfrm_negative_advice;
2431                 if (likely(dst_ops->link_failure == NULL))
2432                         dst_ops->link_failure = xfrm_link_failure;
2433                 if (likely(dst_ops->neigh_lookup == NULL))
2434                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2435                 if (likely(afinfo->garbage_collect == NULL))
2436                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2437                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2438         }
2439         write_unlock_bh(&xfrm_policy_afinfo_lock);
2440
2441         rtnl_lock();
2442         for_each_net(net) {
2443                 struct dst_ops *xfrm_dst_ops;
2444
2445                 switch (afinfo->family) {
2446                 case AF_INET:
2447                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2448                         break;
2449 #if IS_ENABLED(CONFIG_IPV6)
2450                 case AF_INET6:
2451                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2452                         break;
2453 #endif
2454                 default:
2455                         BUG();
2456                 }
2457                 *xfrm_dst_ops = *afinfo->dst_ops;
2458         }
2459         rtnl_unlock();
2460
2461         return err;
2462 }
2463 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2464
2465 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2466 {
2467         int err = 0;
2468         if (unlikely(afinfo == NULL))
2469                 return -EINVAL;
2470         if (unlikely(afinfo->family >= NPROTO))
2471                 return -EAFNOSUPPORT;
2472         write_lock_bh(&xfrm_policy_afinfo_lock);
2473         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2474                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2475                         err = -EINVAL;
2476                 else {
2477                         struct dst_ops *dst_ops = afinfo->dst_ops;
2478                         xfrm_policy_afinfo[afinfo->family] = NULL;
2479                         dst_ops->kmem_cachep = NULL;
2480                         dst_ops->check = NULL;
2481                         dst_ops->negative_advice = NULL;
2482                         dst_ops->link_failure = NULL;
2483                         afinfo->garbage_collect = NULL;
2484                 }
2485         }
2486         write_unlock_bh(&xfrm_policy_afinfo_lock);
2487         return err;
2488 }
2489 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2490
2491 static void __net_init xfrm_dst_ops_init(struct net *net)
2492 {
2493         struct xfrm_policy_afinfo *afinfo;
2494
2495         read_lock_bh(&xfrm_policy_afinfo_lock);
2496         afinfo = xfrm_policy_afinfo[AF_INET];
2497         if (afinfo)
2498                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2499 #if IS_ENABLED(CONFIG_IPV6)
2500         afinfo = xfrm_policy_afinfo[AF_INET6];
2501         if (afinfo)
2502                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2503 #endif
2504         read_unlock_bh(&xfrm_policy_afinfo_lock);
2505 }
2506
2507 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2508 {
2509         struct xfrm_policy_afinfo *afinfo;
2510         if (unlikely(family >= NPROTO))
2511                 return NULL;
2512         read_lock(&xfrm_policy_afinfo_lock);
2513         afinfo = xfrm_policy_afinfo[family];
2514         if (unlikely(!afinfo))
2515                 read_unlock(&xfrm_policy_afinfo_lock);
2516         return afinfo;
2517 }
2518
2519 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2520 {
2521         read_unlock(&xfrm_policy_afinfo_lock);
2522 }
2523
2524 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2525 {
2526         struct net_device *dev = ptr;
2527
2528         switch (event) {
2529         case NETDEV_DOWN:
2530                 xfrm_garbage_collect(dev_net(dev));
2531         }
2532         return NOTIFY_DONE;
2533 }
2534
2535 static struct notifier_block xfrm_dev_notifier = {
2536         .notifier_call  = xfrm_dev_event,
2537 };
2538
2539 #ifdef CONFIG_XFRM_STATISTICS
2540 static int __net_init xfrm_statistics_init(struct net *net)
2541 {
2542         int rv;
2543
2544         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2545                           sizeof(struct linux_xfrm_mib),
2546                           __alignof__(struct linux_xfrm_mib)) < 0)
2547                 return -ENOMEM;
2548         rv = xfrm_proc_init(net);
2549         if (rv < 0)
2550                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2551         return rv;
2552 }
2553
2554 static void xfrm_statistics_fini(struct net *net)
2555 {
2556         xfrm_proc_fini(net);
2557         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2558 }
2559 #else
2560 static int __net_init xfrm_statistics_init(struct net *net)
2561 {
2562         return 0;
2563 }
2564
2565 static void xfrm_statistics_fini(struct net *net)
2566 {
2567 }
2568 #endif
2569
2570 static int __net_init xfrm_policy_init(struct net *net)
2571 {
2572         unsigned int hmask, sz;
2573         int dir;
2574
2575         if (net_eq(net, &init_net))
2576                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2577                                            sizeof(struct xfrm_dst),
2578                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2579                                            NULL);
2580
2581         hmask = 8 - 1;
2582         sz = (hmask+1) * sizeof(struct hlist_head);
2583
2584         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2585         if (!net->xfrm.policy_byidx)
2586                 goto out_byidx;
2587         net->xfrm.policy_idx_hmask = hmask;
2588
2589         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2590                 struct xfrm_policy_hash *htab;
2591
2592                 net->xfrm.policy_count[dir] = 0;
2593                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2594
2595                 htab = &net->xfrm.policy_bydst[dir];
2596                 htab->table = xfrm_hash_alloc(sz);
2597                 if (!htab->table)
2598                         goto out_bydst;
2599                 htab->hmask = hmask;
2600         }
2601
2602         INIT_LIST_HEAD(&net->xfrm.policy_all);
2603         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2604         if (net_eq(net, &init_net))
2605                 register_netdevice_notifier(&xfrm_dev_notifier);
2606         return 0;
2607
2608 out_bydst:
2609         for (dir--; dir >= 0; dir--) {
2610                 struct xfrm_policy_hash *htab;
2611
2612                 htab = &net->xfrm.policy_bydst[dir];
2613                 xfrm_hash_free(htab->table, sz);
2614         }
2615         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2616 out_byidx:
2617         return -ENOMEM;
2618 }
2619
2620 static void xfrm_policy_fini(struct net *net)
2621 {
2622         struct xfrm_audit audit_info;
2623         unsigned int sz;
2624         int dir;
2625
2626         flush_work(&net->xfrm.policy_hash_work);
2627 #ifdef CONFIG_XFRM_SUB_POLICY
2628         audit_info.loginuid = -1;
2629         audit_info.sessionid = -1;
2630         audit_info.secid = 0;
2631         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2632 #endif
2633         audit_info.loginuid = -1;
2634         audit_info.sessionid = -1;
2635         audit_info.secid = 0;
2636         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2637
2638         WARN_ON(!list_empty(&net->xfrm.policy_all));
2639
2640         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2641                 struct xfrm_policy_hash *htab;
2642
2643                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2644
2645                 htab = &net->xfrm.policy_bydst[dir];
2646                 sz = (htab->hmask + 1);
2647                 WARN_ON(!hlist_empty(htab->table));
2648                 xfrm_hash_free(htab->table, sz);
2649         }
2650
2651         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2652         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2653         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2654 }
2655
2656 static int __net_init xfrm_net_init(struct net *net)
2657 {
2658         int rv;
2659
2660         rv = xfrm_statistics_init(net);
2661         if (rv < 0)
2662                 goto out_statistics;
2663         rv = xfrm_state_init(net);
2664         if (rv < 0)
2665                 goto out_state;
2666         rv = xfrm_policy_init(net);
2667         if (rv < 0)
2668                 goto out_policy;
2669         xfrm_dst_ops_init(net);
2670         rv = xfrm_sysctl_init(net);
2671         if (rv < 0)
2672                 goto out_sysctl;
2673         return 0;
2674
2675 out_sysctl:
2676         xfrm_policy_fini(net);
2677 out_policy:
2678         xfrm_state_fini(net);
2679 out_state:
2680         xfrm_statistics_fini(net);
2681 out_statistics:
2682         return rv;
2683 }
2684
2685 static void __net_exit xfrm_net_exit(struct net *net)
2686 {
2687         xfrm_sysctl_fini(net);
2688         xfrm_policy_fini(net);
2689         xfrm_state_fini(net);
2690         xfrm_statistics_fini(net);
2691 }
2692
2693 static struct pernet_operations __net_initdata xfrm_net_ops = {
2694         .init = xfrm_net_init,
2695         .exit = xfrm_net_exit,
2696 };
2697
2698 void __init xfrm_init(void)
2699 {
2700         register_pernet_subsys(&xfrm_net_ops);
2701         xfrm_input_init();
2702 }
2703
2704 #ifdef CONFIG_AUDITSYSCALL
2705 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2706                                          struct audit_buffer *audit_buf)
2707 {
2708         struct xfrm_sec_ctx *ctx = xp->security;
2709         struct xfrm_selector *sel = &xp->selector;
2710
2711         if (ctx)
2712                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2713                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2714
2715         switch(sel->family) {
2716         case AF_INET:
2717                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
2718                 if (sel->prefixlen_s != 32)
2719                         audit_log_format(audit_buf, " src_prefixlen=%d",
2720                                          sel->prefixlen_s);
2721                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
2722                 if (sel->prefixlen_d != 32)
2723                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2724                                          sel->prefixlen_d);
2725                 break;
2726         case AF_INET6:
2727                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
2728                 if (sel->prefixlen_s != 128)
2729                         audit_log_format(audit_buf, " src_prefixlen=%d",
2730                                          sel->prefixlen_s);
2731                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
2732                 if (sel->prefixlen_d != 128)
2733                         audit_log_format(audit_buf, " dst_prefixlen=%d",
2734                                          sel->prefixlen_d);
2735                 break;
2736         }
2737 }
2738
2739 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
2740                            uid_t auid, u32 sessionid, u32 secid)
2741 {
2742         struct audit_buffer *audit_buf;
2743
2744         audit_buf = xfrm_audit_start("SPD-add");
2745         if (audit_buf == NULL)
2746                 return;
2747         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2748         audit_log_format(audit_buf, " res=%u", result);
2749         xfrm_audit_common_policyinfo(xp, audit_buf);
2750         audit_log_end(audit_buf);
2751 }
2752 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
2753
2754 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
2755                               uid_t auid, u32 sessionid, u32 secid)
2756 {
2757         struct audit_buffer *audit_buf;
2758
2759         audit_buf = xfrm_audit_start("SPD-delete");
2760         if (audit_buf == NULL)
2761                 return;
2762         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
2763         audit_log_format(audit_buf, " res=%u", result);
2764         xfrm_audit_common_policyinfo(xp, audit_buf);
2765         audit_log_end(audit_buf);
2766 }
2767 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
2768 #endif
2769
2770 #ifdef CONFIG_XFRM_MIGRATE
2771 static int xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
2772                                        const struct xfrm_selector *sel_tgt)
2773 {
2774         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
2775                 if (sel_tgt->family == sel_cmp->family &&
2776                     xfrm_addr_cmp(&sel_tgt->daddr, &sel_cmp->daddr,
2777                                   sel_cmp->family) == 0 &&
2778                     xfrm_addr_cmp(&sel_tgt->saddr, &sel_cmp->saddr,
2779                                   sel_cmp->family) == 0 &&
2780                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
2781                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
2782                         return 1;
2783                 }
2784         } else {
2785                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
2786                         return 1;
2787                 }
2788         }
2789         return 0;
2790 }
2791
2792 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
2793                                                      u8 dir, u8 type)
2794 {
2795         struct xfrm_policy *pol, *ret = NULL;
2796         struct hlist_node *entry;
2797         struct hlist_head *chain;
2798         u32 priority = ~0U;
2799
2800         read_lock_bh(&xfrm_policy_lock);
2801         chain = policy_hash_direct(&init_net, &sel->daddr, &sel->saddr, sel->family, dir);
2802         hlist_for_each_entry(pol, entry, chain, bydst) {
2803                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2804                     pol->type == type) {
2805                         ret = pol;
2806                         priority = ret->priority;
2807                         break;
2808                 }
2809         }
2810         chain = &init_net.xfrm.policy_inexact[dir];
2811         hlist_for_each_entry(pol, entry, chain, bydst) {
2812                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
2813                     pol->type == type &&
2814                     pol->priority < priority) {
2815                         ret = pol;
2816                         break;
2817                 }
2818         }
2819
2820         if (ret)
2821                 xfrm_pol_hold(ret);
2822
2823         read_unlock_bh(&xfrm_policy_lock);
2824
2825         return ret;
2826 }
2827
2828 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
2829 {
2830         int match = 0;
2831
2832         if (t->mode == m->mode && t->id.proto == m->proto &&
2833             (m->reqid == 0 || t->reqid == m->reqid)) {
2834                 switch (t->mode) {
2835                 case XFRM_MODE_TUNNEL:
2836                 case XFRM_MODE_BEET:
2837                         if (xfrm_addr_cmp(&t->id.daddr, &m->old_daddr,
2838                                           m->old_family) == 0 &&
2839                             xfrm_addr_cmp(&t->saddr, &m->old_saddr,
2840                                           m->old_family) == 0) {
2841                                 match = 1;
2842                         }
2843                         break;
2844                 case XFRM_MODE_TRANSPORT:
2845                         /* in case of transport mode, template does not store
2846                            any IP addresses, hence we just compare mode and
2847                            protocol */
2848                         match = 1;
2849                         break;
2850                 default:
2851                         break;
2852                 }
2853         }
2854         return match;
2855 }
2856
2857 /* update endpoint address(es) of template(s) */
2858 static int xfrm_policy_migrate(struct xfrm_policy *pol,
2859                                struct xfrm_migrate *m, int num_migrate)
2860 {
2861         struct xfrm_migrate *mp;
2862         int i, j, n = 0;
2863
2864         write_lock_bh(&pol->lock);
2865         if (unlikely(pol->walk.dead)) {
2866                 /* target policy has been deleted */
2867                 write_unlock_bh(&pol->lock);
2868                 return -ENOENT;
2869         }
2870
2871         for (i = 0; i < pol->xfrm_nr; i++) {
2872                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
2873                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
2874                                 continue;
2875                         n++;
2876                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
2877                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
2878                                 continue;
2879                         /* update endpoints */
2880                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
2881                                sizeof(pol->xfrm_vec[i].id.daddr));
2882                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
2883                                sizeof(pol->xfrm_vec[i].saddr));
2884                         pol->xfrm_vec[i].encap_family = mp->new_family;
2885                         /* flush bundles */
2886                         atomic_inc(&pol->genid);
2887                 }
2888         }
2889
2890         write_unlock_bh(&pol->lock);
2891
2892         if (!n)
2893                 return -ENODATA;
2894
2895         return 0;
2896 }
2897
2898 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
2899 {
2900         int i, j;
2901
2902         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
2903                 return -EINVAL;
2904
2905         for (i = 0; i < num_migrate; i++) {
2906                 if ((xfrm_addr_cmp(&m[i].old_daddr, &m[i].new_daddr,
2907                                    m[i].old_family) == 0) &&
2908                     (xfrm_addr_cmp(&m[i].old_saddr, &m[i].new_saddr,
2909                                    m[i].old_family) == 0))
2910                         return -EINVAL;
2911                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
2912                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
2913                         return -EINVAL;
2914
2915                 /* check if there is any duplicated entry */
2916                 for (j = i + 1; j < num_migrate; j++) {
2917                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
2918                                     sizeof(m[i].old_daddr)) &&
2919                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
2920                                     sizeof(m[i].old_saddr)) &&
2921                             m[i].proto == m[j].proto &&
2922                             m[i].mode == m[j].mode &&
2923                             m[i].reqid == m[j].reqid &&
2924                             m[i].old_family == m[j].old_family)
2925                                 return -EINVAL;
2926                 }
2927         }
2928
2929         return 0;
2930 }
2931
2932 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
2933                  struct xfrm_migrate *m, int num_migrate,
2934                  struct xfrm_kmaddress *k)
2935 {
2936         int i, err, nx_cur = 0, nx_new = 0;
2937         struct xfrm_policy *pol = NULL;
2938         struct xfrm_state *x, *xc;
2939         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
2940         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
2941         struct xfrm_migrate *mp;
2942
2943         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
2944                 goto out;
2945
2946         /* Stage 1 - find policy */
2947         if ((pol = xfrm_migrate_policy_find(sel, dir, type)) == NULL) {
2948                 err = -ENOENT;
2949                 goto out;
2950         }
2951
2952         /* Stage 2 - find and update state(s) */
2953         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
2954                 if ((x = xfrm_migrate_state_find(mp))) {
2955                         x_cur[nx_cur] = x;
2956                         nx_cur++;
2957                         if ((xc = xfrm_state_migrate(x, mp))) {
2958                                 x_new[nx_new] = xc;
2959                                 nx_new++;
2960                         } else {
2961                                 err = -ENODATA;
2962                                 goto restore_state;
2963                         }
2964                 }
2965         }
2966
2967         /* Stage 3 - update policy */
2968         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
2969                 goto restore_state;
2970
2971         /* Stage 4 - delete old state(s) */
2972         if (nx_cur) {
2973                 xfrm_states_put(x_cur, nx_cur);
2974                 xfrm_states_delete(x_cur, nx_cur);
2975         }
2976
2977         /* Stage 5 - announce */
2978         km_migrate(sel, dir, type, m, num_migrate, k);
2979
2980         xfrm_pol_put(pol);
2981
2982         return 0;
2983 out:
2984         return err;
2985
2986 restore_state:
2987         if (pol)
2988                 xfrm_pol_put(pol);
2989         if (nx_cur)
2990                 xfrm_states_put(x_cur, nx_cur);
2991         if (nx_new)
2992                 xfrm_states_delete(x_new, nx_new);
2993
2994         return err;
2995 }
2996 EXPORT_SYMBOL(xfrm_migrate);
2997 #endif