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 XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  40 #define XFRM_MAX_QUEUE_LEN      100
  41
  42 static struct dst_entry *xfrm_policy_sk_bundles;
  43
  44 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
  45 static struct xfrm_policy_afinfo __rcu *xfrm_policy_afinfo[NPROTO]
  46                                                 __read_mostly;
  47
  48 static struct kmem_cache *xfrm_dst_cache __read_mostly;
  49
  50 static void xfrm_init_pmtu(struct dst_entry *dst);
  51 static int stale_bundle(struct dst_entry *dst);
  52 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
  53 static void xfrm_policy_queue_process(unsigned long arg);
  54
  55 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
  56                                                 int dir);
  57
  58 static inline bool
  59 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  60 {
  61         const struct flowi4 *fl4 = &fl->u.ip4;
  62
  63         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
  64                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
  65                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
  66                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
  67                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
  68                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
  69 }
  70
  71 static inline bool
  72 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
  73 {
  74         const struct flowi6 *fl6 = &fl->u.ip6;
  75
  76         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
  77                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
  78                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
  79                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
  80                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
  81                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
  82 }
  83
  84 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
  85                          unsigned short family)
  86 {
  87         switch (family) {
  88         case AF_INET:
  89                 return __xfrm4_selector_match(sel, fl);
  90         case AF_INET6:
  91                 return __xfrm6_selector_match(sel, fl);
  92         }
  93         return false;
  94 }
  95
  96 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
  97 {
  98         struct xfrm_policy_afinfo *afinfo;
  99
 100         if (unlikely(family >= NPROTO))
 101                 return NULL;
 102         rcu_read_lock();
 103         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 104         if (unlikely(!afinfo))
 105                 rcu_read_unlock();
 106         return afinfo;
 107 }
 108
 109 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
 110 {
 111         rcu_read_unlock();
 112 }
 113
 114 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
 115                                                   const xfrm_address_t *saddr,
 116                                                   const xfrm_address_t *daddr,
 117                                                   int family)
 118 {
 119         struct xfrm_policy_afinfo *afinfo;
 120         struct dst_entry *dst;
 121
 122         afinfo = xfrm_policy_get_afinfo(family);
 123         if (unlikely(afinfo == NULL))
 124                 return ERR_PTR(-EAFNOSUPPORT);
 125
 126         dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 127
 128         xfrm_policy_put_afinfo(afinfo);
 129
 130         return dst;
 131 }
 132
 133 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
 134                                                 xfrm_address_t *prev_saddr,
 135                                                 xfrm_address_t *prev_daddr,
 136                                                 int family)
 137 {
 138         struct net *net = xs_net(x);
 139         xfrm_address_t *saddr = &x->props.saddr;
 140         xfrm_address_t *daddr = &x->id.daddr;
 141         struct dst_entry *dst;
 142
 143         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 144                 saddr = x->coaddr;
 145                 daddr = prev_daddr;
 146         }
 147         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 148                 saddr = prev_saddr;
 149                 daddr = x->coaddr;
 150         }
 151
 152         dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 153
 154         if (!IS_ERR(dst)) {
 155                 if (prev_saddr != saddr)
 156                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 157                 if (prev_daddr != daddr)
 158                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 159         }
 160
 161         return dst;
 162 }
 163
 164 static inline unsigned long make_jiffies(long secs)
 165 {
 166         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 167                 return MAX_SCHEDULE_TIMEOUT-1;
 168         else
 169                 return secs*HZ;
 170 }
 171
 172 static void xfrm_policy_timer(unsigned long data)
 173 {
 174         struct xfrm_policy *xp = (struct xfrm_policy*)data;
 175         unsigned long now = get_seconds();
 176         long next = LONG_MAX;
 177         int warn = 0;
 178         int dir;
 179
 180         read_lock(&xp->lock);
 181
 182         if (unlikely(xp->walk.dead))
 183                 goto out;
 184
 185         dir = xfrm_policy_id2dir(xp->index);
 186
 187         if (xp->lft.hard_add_expires_seconds) {
 188                 long tmo = xp->lft.hard_add_expires_seconds +
 189                         xp->curlft.add_time - now;
 190                 if (tmo <= 0)
 191                         goto expired;
 192                 if (tmo < next)
 193                         next = tmo;
 194         }
 195         if (xp->lft.hard_use_expires_seconds) {
 196                 long tmo = xp->lft.hard_use_expires_seconds +
 197                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 198                 if (tmo <= 0)
 199                         goto expired;
 200                 if (tmo < next)
 201                         next = tmo;
 202         }
 203         if (xp->lft.soft_add_expires_seconds) {
 204                 long tmo = xp->lft.soft_add_expires_seconds +
 205                         xp->curlft.add_time - now;
 206                 if (tmo <= 0) {
 207                         warn = 1;
 208                         tmo = XFRM_KM_TIMEOUT;
 209                 }
 210                 if (tmo < next)
 211                         next = tmo;
 212         }
 213         if (xp->lft.soft_use_expires_seconds) {
 214                 long tmo = xp->lft.soft_use_expires_seconds +
 215                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 216                 if (tmo <= 0) {
 217                         warn = 1;
 218                         tmo = XFRM_KM_TIMEOUT;
 219                 }
 220                 if (tmo < next)
 221                         next = tmo;
 222         }
 223
 224         if (warn)
 225                 km_policy_expired(xp, dir, 0, 0);
 226         if (next != LONG_MAX &&
 227             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 228                 xfrm_pol_hold(xp);
 229
 230 out:
 231         read_unlock(&xp->lock);
 232         xfrm_pol_put(xp);
 233         return;
 234
 235 expired:
 236         read_unlock(&xp->lock);
 237         if (!xfrm_policy_delete(xp, dir))
 238                 km_policy_expired(xp, dir, 1, 0);
 239         xfrm_pol_put(xp);
 240 }
 241
 242 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo)
 243 {
 244         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 245
 246         if (unlikely(pol->walk.dead))
 247                 flo = NULL;
 248         else
 249                 xfrm_pol_hold(pol);
 250
 251         return flo;
 252 }
 253
 254 static int xfrm_policy_flo_check(struct flow_cache_object *flo)
 255 {
 256         struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo);
 257
 258         return !pol->walk.dead;
 259 }
 260
 261 static void xfrm_policy_flo_delete(struct flow_cache_object *flo)
 262 {
 263         xfrm_pol_put(container_of(flo, struct xfrm_policy, flo));
 264 }
 265
 266 static const struct flow_cache_ops xfrm_policy_fc_ops = {
 267         .get = xfrm_policy_flo_get,
 268         .check = xfrm_policy_flo_check,
 269         .delete = xfrm_policy_flo_delete,
 270 };
 271
 272 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 273  * SPD calls.
 274  */
 275
 276 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 277 {
 278         struct xfrm_policy *policy;
 279
 280         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 281
 282         if (policy) {
 283                 write_pnet(&policy->xp_net, net);
 284                 INIT_LIST_HEAD(&policy->walk.all);
 285                 INIT_HLIST_NODE(&policy->bydst);
 286                 INIT_HLIST_NODE(&policy->byidx);
 287                 rwlock_init(&policy->lock);
 288                 atomic_set(&policy->refcnt, 1);
 289                 skb_queue_head_init(&policy->polq.hold_queue);
 290                 setup_timer(&policy->timer, xfrm_policy_timer,
 291                                 (unsigned long)policy);
 292                 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process,
 293                             (unsigned long)policy);
 294                 policy->flo.ops = &xfrm_policy_fc_ops;
 295         }
 296         return policy;
 297 }
 298 EXPORT_SYMBOL(xfrm_policy_alloc);
 299
 300 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 301
 302 void xfrm_policy_destroy(struct xfrm_policy *policy)
 303 {
 304         BUG_ON(!policy->walk.dead);
 305
 306         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 307                 BUG();
 308
 309         security_xfrm_policy_free(policy->security);
 310         kfree(policy);
 311 }
 312 EXPORT_SYMBOL(xfrm_policy_destroy);
 313
 314 static void xfrm_queue_purge(struct sk_buff_head *list)
 315 {
 316         struct sk_buff *skb;
 317
 318         while ((skb = skb_dequeue(list)) != NULL)
 319                 kfree_skb(skb);
 320 }
 321
 322 /* Rule must be locked. Release descentant resources, announce
 323  * entry dead. The rule must be unlinked from lists to the moment.
 324  */
 325
 326 static void xfrm_policy_kill(struct xfrm_policy *policy)
 327 {
 328         policy->walk.dead = 1;
 329
 330         atomic_inc(&policy->genid);
 331
 332         if (del_timer(&policy->polq.hold_timer))
 333                 xfrm_pol_put(policy);
 334         xfrm_queue_purge(&policy->polq.hold_queue);
 335
 336         if (del_timer(&policy->timer))
 337                 xfrm_pol_put(policy);
 338
 339         xfrm_pol_put(policy);
 340 }
 341
 342 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 343
 344 static inline unsigned int idx_hash(struct net *net, u32 index)
 345 {
 346         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 347 }
 348
 349 static struct hlist_head *policy_hash_bysel(struct net *net,
 350                                             const struct xfrm_selector *sel,
 351                                             unsigned short family, int dir)
 352 {
 353         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 354         unsigned int hash = __sel_hash(sel, family, hmask);
 355
 356         return (hash == hmask + 1 ?
 357                 &net->xfrm.policy_inexact[dir] :
 358                 net->xfrm.policy_bydst[dir].table + hash);
 359 }
 360
 361 static struct hlist_head *policy_hash_direct(struct net *net,
 362                                              const xfrm_address_t *daddr,
 363                                              const xfrm_address_t *saddr,
 364                                              unsigned short family, int dir)
 365 {
 366         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 367         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
 368
 369         return net->xfrm.policy_bydst[dir].table + hash;
 370 }
 371
 372 static void xfrm_dst_hash_transfer(struct hlist_head *list,
 373                                    struct hlist_head *ndsttable,
 374                                    unsigned int nhashmask)
 375 {
 376         struct hlist_node *tmp, *entry0 = NULL;
 377         struct xfrm_policy *pol;
 378         unsigned int h0 = 0;
 379
 380 redo:
 381         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 382                 unsigned int h;
 383
 384                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 385                                 pol->family, nhashmask);
 386                 if (!entry0) {
 387                         hlist_del(&pol->bydst);
 388                         hlist_add_head(&pol->bydst, ndsttable+h);
 389                         h0 = h;
 390                 } else {
 391                         if (h != h0)
 392                                 continue;
 393                         hlist_del(&pol->bydst);
 394                         hlist_add_after(entry0, &pol->bydst);
 395                 }
 396                 entry0 = &pol->bydst;
 397         }
 398         if (!hlist_empty(list)) {
 399                 entry0 = NULL;
 400                 goto redo;
 401         }
 402 }
 403
 404 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 405                                    struct hlist_head *nidxtable,
 406                                    unsigned int nhashmask)
 407 {
 408         struct hlist_node *tmp;
 409         struct xfrm_policy *pol;
 410
 411         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 412                 unsigned int h;
 413
 414                 h = __idx_hash(pol->index, nhashmask);
 415                 hlist_add_head(&pol->byidx, nidxtable+h);
 416         }
 417 }
 418
 419 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 420 {
 421         return ((old_hmask + 1) << 1) - 1;
 422 }
 423
 424 static void xfrm_bydst_resize(struct net *net, int dir)
 425 {
 426         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 427         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 428         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 429         struct hlist_head *odst = net->xfrm.policy_bydst[dir].table;
 430         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 431         int i;
 432
 433         if (!ndst)
 434                 return;
 435
 436         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 437
 438         for (i = hmask; i >= 0; i--)
 439                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
 440
 441         net->xfrm.policy_bydst[dir].table = ndst;
 442         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 443
 444         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 445
 446         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 447 }
 448
 449 static void xfrm_byidx_resize(struct net *net, int total)
 450 {
 451         unsigned int hmask = net->xfrm.policy_idx_hmask;
 452         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 453         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 454         struct hlist_head *oidx = net->xfrm.policy_byidx;
 455         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 456         int i;
 457
 458         if (!nidx)
 459                 return;
 460
 461         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 462
 463         for (i = hmask; i >= 0; i--)
 464                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 465
 466         net->xfrm.policy_byidx = nidx;
 467         net->xfrm.policy_idx_hmask = nhashmask;
 468
 469         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 470
 471         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 472 }
 473
 474 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 475 {
 476         unsigned int cnt = net->xfrm.policy_count[dir];
 477         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 478
 479         if (total)
 480                 *total += cnt;
 481
 482         if ((hmask + 1) < xfrm_policy_hashmax &&
 483             cnt > hmask)
 484                 return 1;
 485
 486         return 0;
 487 }
 488
 489 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 490 {
 491         unsigned int hmask = net->xfrm.policy_idx_hmask;
 492
 493         if ((hmask + 1) < xfrm_policy_hashmax &&
 494             total > hmask)
 495                 return 1;
 496
 497         return 0;
 498 }
 499
 500 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 501 {
 502         read_lock_bh(&net->xfrm.xfrm_policy_lock);
 503         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 504         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 505         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 506         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 507         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 508         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 509         si->spdhcnt = net->xfrm.policy_idx_hmask;
 510         si->spdhmcnt = xfrm_policy_hashmax;
 511         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
 512 }
 513 EXPORT_SYMBOL(xfrm_spd_getinfo);
 514
 515 static DEFINE_MUTEX(hash_resize_mutex);
 516 static void xfrm_hash_resize(struct work_struct *work)
 517 {
 518         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 519         int dir, total;
 520
 521         mutex_lock(&hash_resize_mutex);
 522
 523         total = 0;
 524         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
 525                 if (xfrm_bydst_should_resize(net, dir, &total))
 526                         xfrm_bydst_resize(net, dir);
 527         }
 528         if (xfrm_byidx_should_resize(net, total))
 529                 xfrm_byidx_resize(net, total);
 530
 531         mutex_unlock(&hash_resize_mutex);
 532 }
 533
 534 /* Generate new index... KAME seems to generate them ordered by cost
 535  * of an absolute inpredictability of ordering of rules. This will not pass. */
 536 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
 537 {
 538         static u32 idx_generator;
 539
 540         for (;;) {
 541                 struct hlist_head *list;
 542                 struct xfrm_policy *p;
 543                 u32 idx;
 544                 int found;
 545
 546                 if (!index) {
 547                         idx = (idx_generator | dir);
 548                         idx_generator += 8;
 549                 } else {
 550                         idx = index;
 551                         index = 0;
 552                 }
 553
 554                 if (idx == 0)
 555                         idx = 8;
 556                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
 557                 found = 0;
 558                 hlist_for_each_entry(p, list, byidx) {
 559                         if (p->index == idx) {
 560                                 found = 1;
 561                                 break;
 562                         }
 563                 }
 564                 if (!found)
 565                         return idx;
 566         }
 567 }
 568
 569 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
 570 {
 571         u32 *p1 = (u32 *) s1;
 572         u32 *p2 = (u32 *) s2;
 573         int len = sizeof(struct xfrm_selector) / sizeof(u32);
 574         int i;
 575
 576         for (i = 0; i < len; i++) {
 577                 if (p1[i] != p2[i])
 578                         return 1;
 579         }
 580
 581         return 0;
 582 }
 583
 584 static void xfrm_policy_requeue(struct xfrm_policy *old,
 585                                 struct xfrm_policy *new)
 586 {
 587         struct xfrm_policy_queue *pq = &old->polq;
 588         struct sk_buff_head list;
 589
 590         __skb_queue_head_init(&list);
 591
 592         spin_lock_bh(&pq->hold_queue.lock);
 593         skb_queue_splice_init(&pq->hold_queue, &list);
 594         if (del_timer(&pq->hold_timer))
 595                 xfrm_pol_put(old);
 596         spin_unlock_bh(&pq->hold_queue.lock);
 597
 598         if (skb_queue_empty(&list))
 599                 return;
 600
 601         pq = &new->polq;
 602
 603         spin_lock_bh(&pq->hold_queue.lock);
 604         skb_queue_splice(&list, &pq->hold_queue);
 605         pq->timeout = XFRM_QUEUE_TMO_MIN;
 606         if (!mod_timer(&pq->hold_timer, jiffies))
 607                 xfrm_pol_hold(new);
 608         spin_unlock_bh(&pq->hold_queue.lock);
 609 }
 610
 611 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
 612                                    struct xfrm_policy *pol)
 613 {
 614         u32 mark = policy->mark.v & policy->mark.m;
 615
 616         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
 617                 return true;
 618
 619         if ((mark & pol->mark.m) == pol->mark.v &&
 620             policy->priority == pol->priority)
 621                 return true;
 622
 623         return false;
 624 }
 625
 626 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 627 {
 628         struct net *net = xp_net(policy);
 629         struct xfrm_policy *pol;
 630         struct xfrm_policy *delpol;
 631         struct hlist_head *chain;
 632         struct hlist_node *newpos;
 633
 634         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 635         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
 636         delpol = NULL;
 637         newpos = NULL;
 638         hlist_for_each_entry(pol, chain, bydst) {
 639                 if (pol->type == policy->type &&
 640                     !selector_cmp(&pol->selector, &policy->selector) &&
 641                     xfrm_policy_mark_match(policy, pol) &&
 642                     xfrm_sec_ctx_match(pol->security, policy->security) &&
 643                     !WARN_ON(delpol)) {
 644                         if (excl) {
 645                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 646                                 return -EEXIST;
 647                         }
 648                         delpol = pol;
 649                         if (policy->priority > pol->priority)
 650                                 continue;
 651                 } else if (policy->priority >= pol->priority) {
 652                         newpos = &pol->bydst;
 653                         continue;
 654                 }
 655                 if (delpol)
 656                         break;
 657         }
 658         if (newpos)
 659                 hlist_add_after(newpos, &policy->bydst);
 660         else
 661                 hlist_add_head(&policy->bydst, chain);
 662         xfrm_pol_hold(policy);
 663         net->xfrm.policy_count[dir]++;
 664         atomic_inc(&flow_cache_genid);
 665
 666         /* After previous checking, family can either be AF_INET or AF_INET6 */
 667         if (policy->family == AF_INET)
 668                 rt_genid_bump_ipv4(net);
 669         else
 670                 rt_genid_bump_ipv6(net);
 671
 672         if (delpol) {
 673                 xfrm_policy_requeue(delpol, policy);
 674                 __xfrm_policy_unlink(delpol, dir);
 675         }
 676         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
 677         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
 678         policy->curlft.add_time = get_seconds();
 679         policy->curlft.use_time = 0;
 680         if (!mod_timer(&policy->timer, jiffies + HZ))
 681                 xfrm_pol_hold(policy);
 682         list_add(&policy->walk.all, &net->xfrm.policy_all);
 683         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 684
 685         if (delpol)
 686                 xfrm_policy_kill(delpol);
 687         else if (xfrm_bydst_should_resize(net, dir, NULL))
 688                 schedule_work(&net->xfrm.policy_hash_work);
 689
 690         return 0;
 691 }
 692 EXPORT_SYMBOL(xfrm_policy_insert);
 693
 694 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type,
 695                                           int dir, struct xfrm_selector *sel,
 696                                           struct xfrm_sec_ctx *ctx, int delete,
 697                                           int *err)
 698 {
 699         struct xfrm_policy *pol, *ret;
 700         struct hlist_head *chain;
 701
 702         *err = 0;
 703         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 704         chain = policy_hash_bysel(net, sel, sel->family, dir);
 705         ret = NULL;
 706         hlist_for_each_entry(pol, chain, bydst) {
 707                 if (pol->type == type &&
 708                     (mark & pol->mark.m) == pol->mark.v &&
 709                     !selector_cmp(sel, &pol->selector) &&
 710                     xfrm_sec_ctx_match(ctx, pol->security)) {
 711                         xfrm_pol_hold(pol);
 712                         if (delete) {
 713                                 *err = security_xfrm_policy_delete(
 714                                                                 pol->security);
 715                                 if (*err) {
 716                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 717                                         return pol;
 718                                 }
 719                                 __xfrm_policy_unlink(pol, dir);
 720                         }
 721                         ret = pol;
 722                         break;
 723                 }
 724         }
 725         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 726
 727         if (ret && delete)
 728                 xfrm_policy_kill(ret);
 729         return ret;
 730 }
 731 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
 732
 733 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type,
 734                                      int dir, u32 id, int delete, int *err)
 735 {
 736         struct xfrm_policy *pol, *ret;
 737         struct hlist_head *chain;
 738
 739         *err = -ENOENT;
 740         if (xfrm_policy_id2dir(id) != dir)
 741                 return NULL;
 742
 743         *err = 0;
 744         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 745         chain = net->xfrm.policy_byidx + idx_hash(net, id);
 746         ret = NULL;
 747         hlist_for_each_entry(pol, chain, byidx) {
 748                 if (pol->type == type && pol->index == id &&
 749                     (mark & pol->mark.m) == pol->mark.v) {
 750                         xfrm_pol_hold(pol);
 751                         if (delete) {
 752                                 *err = security_xfrm_policy_delete(
 753                                                                 pol->security);
 754                                 if (*err) {
 755                                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 756                                         return pol;
 757                                 }
 758                                 __xfrm_policy_unlink(pol, dir);
 759                         }
 760                         ret = pol;
 761                         break;
 762                 }
 763         }
 764         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 765
 766         if (ret && delete)
 767                 xfrm_policy_kill(ret);
 768         return ret;
 769 }
 770 EXPORT_SYMBOL(xfrm_policy_byid);
 771
 772 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 773 static inline int
 774 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 775 {
 776         int dir, err = 0;
 777
 778         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 779                 struct xfrm_policy *pol;
 780                 int i;
 781
 782                 hlist_for_each_entry(pol,
 783                                      &net->xfrm.policy_inexact[dir], bydst) {
 784                         if (pol->type != type)
 785                                 continue;
 786                         err = security_xfrm_policy_delete(pol->security);
 787                         if (err) {
 788                                 xfrm_audit_policy_delete(pol, 0,
 789                                                          audit_info->loginuid,
 790                                                          audit_info->sessionid,
 791                                                          audit_info->secid);
 792                                 return err;
 793                         }
 794                 }
 795                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 796                         hlist_for_each_entry(pol,
 797                                              net->xfrm.policy_bydst[dir].table + i,
 798                                              bydst) {
 799                                 if (pol->type != type)
 800                                         continue;
 801                                 err = security_xfrm_policy_delete(
 802                                                                 pol->security);
 803                                 if (err) {
 804                                         xfrm_audit_policy_delete(pol, 0,
 805                                                         audit_info->loginuid,
 806                                                         audit_info->sessionid,
 807                                                         audit_info->secid);
 808                                         return err;
 809                                 }
 810                         }
 811                 }
 812         }
 813         return err;
 814 }
 815 #else
 816 static inline int
 817 xfrm_policy_flush_secctx_check(struct net *net, u8 type, struct xfrm_audit *audit_info)
 818 {
 819         return 0;
 820 }
 821 #endif
 822
 823 int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
 824 {
 825         int dir, err = 0, cnt = 0;
 826
 827         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 828
 829         err = xfrm_policy_flush_secctx_check(net, type, audit_info);
 830         if (err)
 831                 goto out;
 832
 833         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 834                 struct xfrm_policy *pol;
 835                 int i;
 836
 837         again1:
 838                 hlist_for_each_entry(pol,
 839                                      &net->xfrm.policy_inexact[dir], bydst) {
 840                         if (pol->type != type)
 841                                 continue;
 842                         __xfrm_policy_unlink(pol, dir);
 843                         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 844                         cnt++;
 845
 846                         xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
 847                                                  audit_info->sessionid,
 848                                                  audit_info->secid);
 849
 850                         xfrm_policy_kill(pol);
 851
 852                         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 853                         goto again1;
 854                 }
 855
 856                 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
 857         again2:
 858                         hlist_for_each_entry(pol,
 859                                              net->xfrm.policy_bydst[dir].table + i,
 860                                              bydst) {
 861                                 if (pol->type != type)
 862                                         continue;
 863                                 __xfrm_policy_unlink(pol, dir);
 864                                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 865                                 cnt++;
 866
 867                                 xfrm_audit_policy_delete(pol, 1,
 868                                                          audit_info->loginuid,
 869                                                          audit_info->sessionid,
 870                                                          audit_info->secid);
 871                                 xfrm_policy_kill(pol);
 872
 873                                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
 874                                 goto again2;
 875                         }
 876                 }
 877
 878         }
 879         if (!cnt)
 880                 err = -ESRCH;
 881 out:
 882         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 883         return err;
 884 }
 885 EXPORT_SYMBOL(xfrm_policy_flush);
 886
 887 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
 888                      int (*func)(struct xfrm_policy *, int, int, void*),
 889                      void *data)
 890 {
 891         struct xfrm_policy *pol;
 892         struct xfrm_policy_walk_entry *x;
 893         int error = 0;
 894
 895         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
 896             walk->type != XFRM_POLICY_TYPE_ANY)
 897                 return -EINVAL;
 898
 899         if (list_empty(&walk->walk.all) && walk->seq != 0)
 900                 return 0;
 901
 902         write_lock_bh(&net->xfrm.xfrm_policy_lock);
 903         if (list_empty(&walk->walk.all))
 904                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
 905         else
 906                 x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
 907         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
 908                 if (x->dead)
 909                         continue;
 910                 pol = container_of(x, struct xfrm_policy, walk);
 911                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
 912                     walk->type != pol->type)
 913                         continue;
 914                 error = func(pol, xfrm_policy_id2dir(pol->index),
 915                              walk->seq, data);
 916                 if (error) {
 917                         list_move_tail(&walk->walk.all, &x->all);
 918                         goto out;
 919                 }
 920                 walk->seq++;
 921         }
 922         if (walk->seq == 0) {
 923                 error = -ENOENT;
 924                 goto out;
 925         }
 926         list_del_init(&walk->walk.all);
 927 out:
 928         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 929         return error;
 930 }
 931 EXPORT_SYMBOL(xfrm_policy_walk);
 932
 933 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
 934 {
 935         INIT_LIST_HEAD(&walk->walk.all);
 936         walk->walk.dead = 1;
 937         walk->type = type;
 938         walk->seq = 0;
 939 }
 940 EXPORT_SYMBOL(xfrm_policy_walk_init);
 941
 942 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
 943 {
 944         if (list_empty(&walk->walk.all))
 945                 return;
 946
 947         write_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
 948         list_del(&walk->walk.all);
 949         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
 950 }
 951 EXPORT_SYMBOL(xfrm_policy_walk_done);
 952
 953 /*
 954  * Find policy to apply to this flow.
 955  *
 956  * Returns 0 if policy found, else an -errno.
 957  */
 958 static int xfrm_policy_match(const struct xfrm_policy *pol,
 959                              const struct flowi *fl,
 960                              u8 type, u16 family, int dir)
 961 {
 962         const struct xfrm_selector *sel = &pol->selector;
 963         int ret = -ESRCH;
 964         bool match;
 965
 966         if (pol->family != family ||
 967             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
 968             pol->type != type)
 969                 return ret;
 970
 971         match = xfrm_selector_match(sel, fl, family);
 972         if (match)
 973                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
 974                                                   dir);
 975
 976         return ret;
 977 }
 978
 979 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
 980                                                      const struct flowi *fl,
 981                                                      u16 family, u8 dir)
 982 {
 983         int err;
 984         struct xfrm_policy *pol, *ret;
 985         const xfrm_address_t *daddr, *saddr;
 986         struct hlist_head *chain;
 987         u32 priority = ~0U;
 988
 989         daddr = xfrm_flowi_daddr(fl, family);
 990         saddr = xfrm_flowi_saddr(fl, family);
 991         if (unlikely(!daddr || !saddr))
 992                 return NULL;
 993
 994         read_lock_bh(&net->xfrm.xfrm_policy_lock);
 995         chain = policy_hash_direct(net, daddr, saddr, family, dir);
 996         ret = NULL;
 997         hlist_for_each_entry(pol, chain, bydst) {
 998                 err = xfrm_policy_match(pol, fl, type, family, dir);
 999                 if (err) {
1000                         if (err == -ESRCH)
1001                                 continue;
1002                         else {
1003                                 ret = ERR_PTR(err);
1004                                 goto fail;
1005                         }
1006                 } else {
1007                         ret = pol;
1008                         priority = ret->priority;
1009                         break;
1010                 }
1011         }
1012         chain = &net->xfrm.policy_inexact[dir];
1013         hlist_for_each_entry(pol, chain, bydst) {
1014                 err = xfrm_policy_match(pol, fl, type, family, dir);
1015                 if (err) {
1016                         if (err == -ESRCH)
1017                                 continue;
1018                         else {
1019                                 ret = ERR_PTR(err);
1020                                 goto fail;
1021                         }
1022                 } else if (pol->priority < priority) {
1023                         ret = pol;
1024                         break;
1025                 }
1026         }
1027         if (ret)
1028                 xfrm_pol_hold(ret);
1029 fail:
1030         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1031
1032         return ret;
1033 }
1034
1035 static struct xfrm_policy *
1036 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir)
1037 {
1038 #ifdef CONFIG_XFRM_SUB_POLICY
1039         struct xfrm_policy *pol;
1040
1041         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
1042         if (pol != NULL)
1043                 return pol;
1044 #endif
1045         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1046 }
1047
1048 static int flow_to_policy_dir(int dir)
1049 {
1050         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1051             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1052             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1053                 return dir;
1054
1055         switch (dir) {
1056         default:
1057         case FLOW_DIR_IN:
1058                 return XFRM_POLICY_IN;
1059         case FLOW_DIR_OUT:
1060                 return XFRM_POLICY_OUT;
1061         case FLOW_DIR_FWD:
1062                 return XFRM_POLICY_FWD;
1063         }
1064 }
1065
1066 static struct flow_cache_object *
1067 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family,
1068                    u8 dir, struct flow_cache_object *old_obj, void *ctx)
1069 {
1070         struct xfrm_policy *pol;
1071
1072         if (old_obj)
1073                 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo));
1074
1075         pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir));
1076         if (IS_ERR_OR_NULL(pol))
1077                 return ERR_CAST(pol);
1078
1079         /* Resolver returns two references:
1080          * one for cache and one for caller of flow_cache_lookup() */
1081         xfrm_pol_hold(pol);
1082
1083         return &pol->flo;
1084 }
1085
1086 static inline int policy_to_flow_dir(int dir)
1087 {
1088         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1089             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1090             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1091                 return dir;
1092         switch (dir) {
1093         default:
1094         case XFRM_POLICY_IN:
1095                 return FLOW_DIR_IN;
1096         case XFRM_POLICY_OUT:
1097                 return FLOW_DIR_OUT;
1098         case XFRM_POLICY_FWD:
1099                 return FLOW_DIR_FWD;
1100         }
1101 }
1102
1103 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir,
1104                                                  const struct flowi *fl)
1105 {
1106         struct xfrm_policy *pol;
1107         struct net *net = sock_net(sk);
1108
1109         read_lock_bh(&net->xfrm.xfrm_policy_lock);
1110         if ((pol = sk->sk_policy[dir]) != NULL) {
1111                 bool match = xfrm_selector_match(&pol->selector, fl,
1112                                                  sk->sk_family);
1113                 int err = 0;
1114
1115                 if (match) {
1116                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v) {
1117                                 pol = NULL;
1118                                 goto out;
1119                         }
1120                         err = security_xfrm_policy_lookup(pol->security,
1121                                                       fl->flowi_secid,
1122                                                       policy_to_flow_dir(dir));
1123                         if (!err)
1124                                 xfrm_pol_hold(pol);
1125                         else if (err == -ESRCH)
1126                                 pol = NULL;
1127                         else
1128                                 pol = ERR_PTR(err);
1129                 } else
1130                         pol = NULL;
1131         }
1132 out:
1133         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
1134         return pol;
1135 }
1136
1137 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1138 {
1139         struct net *net = xp_net(pol);
1140         struct hlist_head *chain = policy_hash_bysel(net, &pol->selector,
1141                                                      pol->family, dir);
1142
1143         list_add(&pol->walk.all, &net->xfrm.policy_all);
1144         hlist_add_head(&pol->bydst, chain);
1145         hlist_add_head(&pol->byidx, net->xfrm.policy_byidx+idx_hash(net, pol->index));
1146         net->xfrm.policy_count[dir]++;
1147         xfrm_pol_hold(pol);
1148
1149         if (xfrm_bydst_should_resize(net, dir, NULL))
1150                 schedule_work(&net->xfrm.policy_hash_work);
1151 }
1152
1153 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1154                                                 int dir)
1155 {
1156         struct net *net = xp_net(pol);
1157
1158         if (hlist_unhashed(&pol->bydst))
1159                 return NULL;
1160
1161         hlist_del(&pol->bydst);
1162         hlist_del(&pol->byidx);
1163         list_del(&pol->walk.all);
1164         net->xfrm.policy_count[dir]--;
1165
1166         return pol;
1167 }
1168
1169 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1170 {
1171         struct net *net = xp_net(pol);
1172
1173         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1174         pol = __xfrm_policy_unlink(pol, dir);
1175         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1176         if (pol) {
1177                 xfrm_policy_kill(pol);
1178                 return 0;
1179         }
1180         return -ENOENT;
1181 }
1182 EXPORT_SYMBOL(xfrm_policy_delete);
1183
1184 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1185 {
1186         struct net *net = xp_net(pol);
1187         struct xfrm_policy *old_pol;
1188
1189 #ifdef CONFIG_XFRM_SUB_POLICY
1190         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1191                 return -EINVAL;
1192 #endif
1193
1194         write_lock_bh(&net->xfrm.xfrm_policy_lock);
1195         old_pol = sk->sk_policy[dir];
1196         sk->sk_policy[dir] = pol;
1197         if (pol) {
1198                 pol->curlft.add_time = get_seconds();
1199                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
1200                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1201         }
1202         if (old_pol) {
1203                 if (pol)
1204                         xfrm_policy_requeue(old_pol, pol);
1205
1206                 /* Unlinking succeeds always. This is the only function
1207                  * allowed to delete or replace socket policy.
1208                  */
1209                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1210         }
1211         write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1212
1213         if (old_pol) {
1214                 xfrm_policy_kill(old_pol);
1215         }
1216         return 0;
1217 }
1218
1219 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
1220 {
1221         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
1222         struct net *net = xp_net(old);
1223
1224         if (newp) {
1225                 newp->selector = old->selector;
1226                 if (security_xfrm_policy_clone(old->security,
1227                                                &newp->security)) {
1228                         kfree(newp);
1229                         return NULL;  /* ENOMEM */
1230                 }
1231                 newp->lft = old->lft;
1232                 newp->curlft = old->curlft;
1233                 newp->mark = old->mark;
1234                 newp->action = old->action;
1235                 newp->flags = old->flags;
1236                 newp->xfrm_nr = old->xfrm_nr;
1237                 newp->index = old->index;
1238                 newp->type = old->type;
1239                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1240                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1241                 write_lock_bh(&net->xfrm.xfrm_policy_lock);
1242                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1243                 write_unlock_bh(&net->xfrm.xfrm_policy_lock);
1244                 xfrm_pol_put(newp);
1245         }
1246         return newp;
1247 }
1248
1249 int __xfrm_sk_clone_policy(struct sock *sk)
1250 {
1251         struct xfrm_policy *p0 = sk->sk_policy[0],
1252                            *p1 = sk->sk_policy[1];
1253
1254         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1255         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1256                 return -ENOMEM;
1257         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1258                 return -ENOMEM;
1259         return 0;
1260 }
1261
1262 static int
1263 xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
1264                unsigned short family)
1265 {
1266         int err;
1267         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1268
1269         if (unlikely(afinfo == NULL))
1270                 return -EINVAL;
1271         err = afinfo->get_saddr(net, local, remote);
1272         xfrm_policy_put_afinfo(afinfo);
1273         return err;
1274 }
1275
1276 /* Resolve list of templates for the flow, given policy. */
1277
1278 static int
1279 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
1280                       struct xfrm_state **xfrm, unsigned short family)
1281 {
1282         struct net *net = xp_net(policy);
1283         int nx;
1284         int i, error;
1285         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1286         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1287         xfrm_address_t tmp;
1288
1289         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1290                 struct xfrm_state *x;
1291                 xfrm_address_t *remote = daddr;
1292                 xfrm_address_t *local  = saddr;
1293                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1294
1295                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
1296                     tmpl->mode == XFRM_MODE_BEET) {
1297                         remote = &tmpl->id.daddr;
1298                         local = &tmpl->saddr;
1299                         if (xfrm_addr_any(local, tmpl->encap_family)) {
1300                                 error = xfrm_get_saddr(net, &tmp, remote, tmpl->encap_family);
1301                                 if (error)
1302                                         goto fail;
1303                                 local = &tmp;
1304                         }
1305                 }
1306
1307                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1308
1309                 if (x && x->km.state == XFRM_STATE_VALID) {
1310                         xfrm[nx++] = x;
1311                         daddr = remote;
1312                         saddr = local;
1313                         continue;
1314                 }
1315                 if (x) {
1316                         error = (x->km.state == XFRM_STATE_ERROR ?
1317                                  -EINVAL : -EAGAIN);
1318                         xfrm_state_put(x);
1319                 }
1320                 else if (error == -ESRCH)
1321                         error = -EAGAIN;
1322
1323                 if (!tmpl->optional)
1324                         goto fail;
1325         }
1326         return nx;
1327
1328 fail:
1329         for (nx--; nx>=0; nx--)
1330                 xfrm_state_put(xfrm[nx]);
1331         return error;
1332 }
1333
1334 static int
1335 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
1336                   struct xfrm_state **xfrm, unsigned short family)
1337 {
1338         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1339         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1340         int cnx = 0;
1341         int error;
1342         int ret;
1343         int i;
1344
1345         for (i = 0; i < npols; i++) {
1346                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1347                         error = -ENOBUFS;
1348                         goto fail;
1349                 }
1350
1351                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1352                 if (ret < 0) {
1353                         error = ret;
1354                         goto fail;
1355                 } else
1356                         cnx += ret;
1357         }
1358
1359         /* found states are sorted for outbound processing */
1360         if (npols > 1)
1361                 xfrm_state_sort(xfrm, tpp, cnx, family);
1362
1363         return cnx;
1364
1365  fail:
1366         for (cnx--; cnx>=0; cnx--)
1367                 xfrm_state_put(tpp[cnx]);
1368         return error;
1369
1370 }
1371
1372 /* Check that the bundle accepts the flow and its components are
1373  * still valid.
1374  */
1375
1376 static inline int xfrm_get_tos(const struct flowi *fl, int family)
1377 {
1378         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1379         int tos;
1380
1381         if (!afinfo)
1382                 return -EINVAL;
1383
1384         tos = afinfo->get_tos(fl);
1385
1386         xfrm_policy_put_afinfo(afinfo);
1387
1388         return tos;
1389 }
1390
1391 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo)
1392 {
1393         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1394         struct dst_entry *dst = &xdst->u.dst;
1395
1396         if (xdst->route == NULL) {
1397                 /* Dummy bundle - if it has xfrms we were not
1398                  * able to build bundle as template resolution failed.
1399                  * It means we need to try again resolving. */
1400                 if (xdst->num_xfrms > 0)
1401                         return NULL;
1402         } else if (dst->flags & DST_XFRM_QUEUE) {
1403                 return NULL;
1404         } else {
1405                 /* Real bundle */
1406                 if (stale_bundle(dst))
1407                         return NULL;
1408         }
1409
1410         dst_hold(dst);
1411         return flo;
1412 }
1413
1414 static int xfrm_bundle_flo_check(struct flow_cache_object *flo)
1415 {
1416         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1417         struct dst_entry *dst = &xdst->u.dst;
1418
1419         if (!xdst->route)
1420                 return 0;
1421         if (stale_bundle(dst))
1422                 return 0;
1423
1424         return 1;
1425 }
1426
1427 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo)
1428 {
1429         struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo);
1430         struct dst_entry *dst = &xdst->u.dst;
1431
1432         dst_free(dst);
1433 }
1434
1435 static const struct flow_cache_ops xfrm_bundle_fc_ops = {
1436         .get = xfrm_bundle_flo_get,
1437         .check = xfrm_bundle_flo_check,
1438         .delete = xfrm_bundle_flo_delete,
1439 };
1440
1441 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
1442 {
1443         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1444         struct dst_ops *dst_ops;
1445         struct xfrm_dst *xdst;
1446
1447         if (!afinfo)
1448                 return ERR_PTR(-EINVAL);
1449
1450         switch (family) {
1451         case AF_INET:
1452                 dst_ops = &net->xfrm.xfrm4_dst_ops;
1453                 break;
1454 #if IS_ENABLED(CONFIG_IPV6)
1455         case AF_INET6:
1456                 dst_ops = &net->xfrm.xfrm6_dst_ops;
1457                 break;
1458 #endif
1459         default:
1460                 BUG();
1461         }
1462         xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0);
1463
1464         if (likely(xdst)) {
1465                 struct dst_entry *dst = &xdst->u.dst;
1466
1467                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
1468                 xdst->flo.ops = &xfrm_bundle_fc_ops;
1469                 if (afinfo->init_dst)
1470                         afinfo->init_dst(net, xdst);
1471         } else
1472                 xdst = ERR_PTR(-ENOBUFS);
1473
1474         xfrm_policy_put_afinfo(afinfo);
1475
1476         return xdst;
1477 }
1478
1479 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
1480                                  int nfheader_len)
1481 {
1482         struct xfrm_policy_afinfo *afinfo =
1483                 xfrm_policy_get_afinfo(dst->ops->family);
1484         int err;
1485
1486         if (!afinfo)
1487                 return -EINVAL;
1488
1489         err = afinfo->init_path(path, dst, nfheader_len);
1490
1491         xfrm_policy_put_afinfo(afinfo);
1492
1493         return err;
1494 }
1495
1496 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
1497                                 const struct flowi *fl)
1498 {
1499         struct xfrm_policy_afinfo *afinfo =
1500                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
1501         int err;
1502
1503         if (!afinfo)
1504                 return -EINVAL;
1505
1506         err = afinfo->fill_dst(xdst, dev, fl);
1507
1508         xfrm_policy_put_afinfo(afinfo);
1509
1510         return err;
1511 }
1512
1513
1514 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1515  * all the metrics... Shortly, bundle a bundle.
1516  */
1517
1518 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
1519                                             struct xfrm_state **xfrm, int nx,
1520                                             const struct flowi *fl,
1521                                             struct dst_entry *dst)
1522 {
1523         struct net *net = xp_net(policy);
1524         unsigned long now = jiffies;
1525         struct net_device *dev;
1526         struct xfrm_mode *inner_mode;
1527         struct dst_entry *dst_prev = NULL;
1528         struct dst_entry *dst0 = NULL;
1529         int i = 0;
1530         int err;
1531         int header_len = 0;
1532         int nfheader_len = 0;
1533         int trailer_len = 0;
1534         int tos;
1535         int family = policy->selector.family;
1536         xfrm_address_t saddr, daddr;
1537
1538         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
1539
1540         tos = xfrm_get_tos(fl, family);
1541         err = tos;
1542         if (tos < 0)
1543                 goto put_states;
1544
1545         dst_hold(dst);
1546
1547         for (; i < nx; i++) {
1548                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
1549                 struct dst_entry *dst1 = &xdst->u.dst;
1550
1551                 err = PTR_ERR(xdst);
1552                 if (IS_ERR(xdst)) {
1553                         dst_release(dst);
1554                         goto put_states;
1555                 }
1556
1557                 if (xfrm[i]->sel.family == AF_UNSPEC) {
1558                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
1559                                                         xfrm_af2proto(family));
1560                         if (!inner_mode) {
1561                                 err = -EAFNOSUPPORT;
1562                                 dst_release(dst);
1563                                 goto put_states;
1564                         }
1565                 } else
1566                         inner_mode = xfrm[i]->inner_mode;
1567
1568                 if (!dst_prev)
1569                         dst0 = dst1;
1570                 else {
1571                         dst_prev->child = dst_clone(dst1);
1572                         dst1->flags |= DST_NOHASH;
1573                 }
1574
1575                 xdst->route = dst;
1576                 dst_copy_metrics(dst1, dst);
1577
1578                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
1579                         family = xfrm[i]->props.family;
1580                         dst = xfrm_dst_lookup(xfrm[i], tos, &saddr, &daddr,
1581                                               family);
1582                         err = PTR_ERR(dst);
1583                         if (IS_ERR(dst))
1584                                 goto put_states;
1585                 } else
1586                         dst_hold(dst);
1587
1588                 dst1->xfrm = xfrm[i];
1589                 xdst->xfrm_genid = xfrm[i]->genid;
1590
1591                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1592                 dst1->flags |= DST_HOST;
1593                 dst1->lastuse = now;
1594
1595                 dst1->input = dst_discard;
1596                 dst1->output = inner_mode->afinfo->output;
1597
1598                 dst1->next = dst_prev;
1599                 dst_prev = dst1;
1600
1601                 header_len += xfrm[i]->props.header_len;
1602                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
1603                         nfheader_len += xfrm[i]->props.header_len;
1604                 trailer_len += xfrm[i]->props.trailer_len;
1605         }
1606
1607         dst_prev->child = dst;
1608         dst0->path = dst;
1609
1610         err = -ENODEV;
1611         dev = dst->dev;
1612         if (!dev)
1613                 goto free_dst;
1614
1615         xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
1616         xfrm_init_pmtu(dst_prev);
1617
1618         for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) {
1619                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev;
1620
1621                 err = xfrm_fill_dst(xdst, dev, fl);
1622                 if (err)
1623                         goto free_dst;
1624
1625                 dst_prev->header_len = header_len;
1626                 dst_prev->trailer_len = trailer_len;
1627                 header_len -= xdst->u.dst.xfrm->props.header_len;
1628                 trailer_len -= xdst->u.dst.xfrm->props.trailer_len;
1629         }
1630
1631 out:
1632         return dst0;
1633
1634 put_states:
1635         for (; i < nx; i++)
1636                 xfrm_state_put(xfrm[i]);
1637 free_dst:
1638         if (dst0)
1639                 dst_free(dst0);
1640         dst0 = ERR_PTR(err);
1641         goto out;
1642 }
1643
1644 static int inline
1645 xfrm_dst_alloc_copy(void **target, const void *src, int size)
1646 {
1647         if (!*target) {
1648                 *target = kmalloc(size, GFP_ATOMIC);
1649                 if (!*target)
1650                         return -ENOMEM;
1651         }
1652         memcpy(*target, src, size);
1653         return 0;
1654 }
1655
1656 static int inline
1657 xfrm_dst_update_parent(struct dst_entry *dst, const struct xfrm_selector *sel)
1658 {
1659 #ifdef CONFIG_XFRM_SUB_POLICY
1660         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1661         return xfrm_dst_alloc_copy((void **)&(xdst->partner),
1662                                    sel, sizeof(*sel));
1663 #else
1664         return 0;
1665 #endif
1666 }
1667
1668 static int inline
1669 xfrm_dst_update_origin(struct dst_entry *dst, const struct flowi *fl)
1670 {
1671 #ifdef CONFIG_XFRM_SUB_POLICY
1672         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1673         return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl));
1674 #else
1675         return 0;
1676 #endif
1677 }
1678
1679 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
1680                                 struct xfrm_policy **pols,
1681                                 int *num_pols, int *num_xfrms)
1682 {
1683         int i;
1684
1685         if (*num_pols == 0 || !pols[0]) {
1686                 *num_pols = 0;
1687                 *num_xfrms = 0;
1688                 return 0;
1689         }
1690         if (IS_ERR(pols[0]))
1691                 return PTR_ERR(pols[0]);
1692
1693         *num_xfrms = pols[0]->xfrm_nr;
1694
1695 #ifdef CONFIG_XFRM_SUB_POLICY
1696         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
1697             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1698                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
1699                                                     XFRM_POLICY_TYPE_MAIN,
1700                                                     fl, family,
1701                                                     XFRM_POLICY_OUT);
1702                 if (pols[1]) {
1703                         if (IS_ERR(pols[1])) {
1704                                 xfrm_pols_put(pols, *num_pols);
1705                                 return PTR_ERR(pols[1]);
1706                         }
1707                         (*num_pols) ++;
1708                         (*num_xfrms) += pols[1]->xfrm_nr;
1709                 }
1710         }
1711 #endif
1712         for (i = 0; i < *num_pols; i++) {
1713                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
1714                         *num_xfrms = -1;
1715                         break;
1716                 }
1717         }
1718
1719         return 0;
1720
1721 }
1722
1723 static struct xfrm_dst *
1724 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
1725                                const struct flowi *fl, u16 family,
1726                                struct dst_entry *dst_orig)
1727 {
1728         struct net *net = xp_net(pols[0]);
1729         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1730         struct dst_entry *dst;
1731         struct xfrm_dst *xdst;
1732         int err;
1733
1734         /* Try to instantiate a bundle */
1735         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
1736         if (err <= 0) {
1737                 if (err != 0 && err != -EAGAIN)
1738                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
1739                 return ERR_PTR(err);
1740         }
1741
1742         dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig);
1743         if (IS_ERR(dst)) {
1744                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
1745                 return ERR_CAST(dst);
1746         }
1747
1748         xdst = (struct xfrm_dst *)dst;
1749         xdst->num_xfrms = err;
1750         if (num_pols > 1)
1751                 err = xfrm_dst_update_parent(dst, &pols[1]->selector);
1752         else
1753                 err = xfrm_dst_update_origin(dst, fl);
1754         if (unlikely(err)) {
1755                 dst_free(dst);
1756                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
1757                 return ERR_PTR(err);
1758         }
1759
1760         xdst->num_pols = num_pols;
1761         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
1762         xdst->policy_genid = atomic_read(&pols[0]->genid);
1763
1764         return xdst;
1765 }
1766
1767 static void xfrm_policy_queue_process(unsigned long arg)
1768 {
1769         int err = 0;
1770         struct sk_buff *skb;
1771         struct sock *sk;
1772         struct dst_entry *dst;
1773         struct xfrm_policy *pol = (struct xfrm_policy *)arg;
1774         struct xfrm_policy_queue *pq = &pol->polq;
1775         struct flowi fl;
1776         struct sk_buff_head list;
1777
1778         spin_lock(&pq->hold_queue.lock);
1779         skb = skb_peek(&pq->hold_queue);
1780         if (!skb) {
1781                 spin_unlock(&pq->hold_queue.lock);
1782                 goto out;
1783         }
1784         dst = skb_dst(skb);
1785         sk = skb->sk;
1786         xfrm_decode_session(skb, &fl, dst->ops->family);
1787         spin_unlock(&pq->hold_queue.lock);
1788
1789         dst_hold(dst->path);
1790         dst = xfrm_lookup(xp_net(pol), dst->path, &fl,
1791                           sk, 0);
1792         if (IS_ERR(dst))
1793                 goto purge_queue;
1794
1795         if (dst->flags & DST_XFRM_QUEUE) {
1796                 dst_release(dst);
1797
1798                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
1799                         goto purge_queue;
1800
1801                 pq->timeout = pq->timeout << 1;
1802                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
1803                         xfrm_pol_hold(pol);
1804         goto out;
1805         }
1806
1807         dst_release(dst);
1808
1809         __skb_queue_head_init(&list);
1810
1811         spin_lock(&pq->hold_queue.lock);
1812         pq->timeout = 0;
1813         skb_queue_splice_init(&pq->hold_queue, &list);
1814         spin_unlock(&pq->hold_queue.lock);
1815
1816         while (!skb_queue_empty(&list)) {
1817                 skb = __skb_dequeue(&list);
1818
1819                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
1820                 dst_hold(skb_dst(skb)->path);
1821                 dst = xfrm_lookup(xp_net(pol), skb_dst(skb)->path,
1822                                   &fl, skb->sk, 0);
1823                 if (IS_ERR(dst)) {
1824                         kfree_skb(skb);
1825                         continue;
1826                 }
1827
1828                 nf_reset(skb);
1829                 skb_dst_drop(skb);
1830                 skb_dst_set(skb, dst);
1831
1832                 err = dst_output(skb);
1833         }
1834
1835 out:
1836         xfrm_pol_put(pol);
1837         return;
1838
1839 purge_queue:
1840         pq->timeout = 0;
1841         xfrm_queue_purge(&pq->hold_queue);
1842         xfrm_pol_put(pol);
1843 }
1844
1845 static int xdst_queue_output(struct sk_buff *skb)
1846 {
1847         unsigned long sched_next;
1848         struct dst_entry *dst = skb_dst(skb);
1849         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
1850         struct xfrm_policy *pol = xdst->pols[0];
1851         struct xfrm_policy_queue *pq = &pol->polq;
1852         const struct sk_buff *fclone = skb + 1;
1853
1854         if (unlikely(skb->fclone == SKB_FCLONE_ORIG &&
1855                      fclone->fclone == SKB_FCLONE_CLONE)) {
1856                 kfree_skb(skb);
1857                 return 0;
1858         }
1859
1860         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
1861                 kfree_skb(skb);
1862                 return -EAGAIN;
1863         }
1864
1865         skb_dst_force(skb);
1866
1867         spin_lock_bh(&pq->hold_queue.lock);
1868
1869         if (!pq->timeout)
1870                 pq->timeout = XFRM_QUEUE_TMO_MIN;
1871
1872         sched_next = jiffies + pq->timeout;
1873
1874         if (del_timer(&pq->hold_timer)) {
1875                 if (time_before(pq->hold_timer.expires, sched_next))
1876                         sched_next = pq->hold_timer.expires;
1877                 xfrm_pol_put(pol);
1878         }
1879
1880         __skb_queue_tail(&pq->hold_queue, skb);
1881         if (!mod_timer(&pq->hold_timer, sched_next))
1882                 xfrm_pol_hold(pol);
1883
1884         spin_unlock_bh(&pq->hold_queue.lock);
1885
1886         return 0;
1887 }
1888
1889 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
1890                                                  struct dst_entry *dst,
1891                                                  const struct flowi *fl,
1892                                                  int num_xfrms,
1893                                                  u16 family)
1894 {
1895         int err;
1896         struct net_device *dev;
1897         struct dst_entry *dst1;
1898         struct xfrm_dst *xdst;
1899
1900         xdst = xfrm_alloc_dst(net, family);
1901         if (IS_ERR(xdst))
1902                 return xdst;
1903
1904         if (net->xfrm.sysctl_larval_drop || num_xfrms <= 0 ||
1905             (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP))
1906                 return xdst;
1907
1908         dst1 = &xdst->u.dst;
1909         dst_hold(dst);
1910         xdst->route = dst;
1911
1912         dst_copy_metrics(dst1, dst);
1913
1914         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
1915         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
1916         dst1->lastuse = jiffies;
1917
1918         dst1->input = dst_discard;
1919         dst1->output = xdst_queue_output;
1920
1921         dst_hold(dst);
1922         dst1->child = dst;
1923         dst1->path = dst;
1924
1925         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
1926
1927         err = -ENODEV;
1928         dev = dst->dev;
1929         if (!dev)
1930                 goto free_dst;
1931
1932         err = xfrm_fill_dst(xdst, dev, fl);
1933         if (err)
1934                 goto free_dst;
1935
1936 out:
1937         return xdst;
1938
1939 free_dst:
1940         dst_release(dst1);
1941         xdst = ERR_PTR(err);
1942         goto out;
1943 }
1944
1945 static struct flow_cache_object *
1946 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir,
1947                    struct flow_cache_object *oldflo, void *ctx)
1948 {
1949         struct dst_entry *dst_orig = (struct dst_entry *)ctx;
1950         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1951         struct xfrm_dst *xdst, *new_xdst;
1952         int num_pols = 0, num_xfrms = 0, i, err, pol_dead;
1953
1954         /* Check if the policies from old bundle are usable */
1955         xdst = NULL;
1956         if (oldflo) {
1957                 xdst = container_of(oldflo, struct xfrm_dst, flo);
1958                 num_pols = xdst->num_pols;
1959                 num_xfrms = xdst->num_xfrms;
1960                 pol_dead = 0;
1961                 for (i = 0; i < num_pols; i++) {
1962                         pols[i] = xdst->pols[i];
1963                         pol_dead |= pols[i]->walk.dead;
1964                 }
1965                 if (pol_dead) {
1966                         dst_free(&xdst->u.dst);
1967                         xdst = NULL;
1968                         num_pols = 0;
1969                         num_xfrms = 0;
1970                         oldflo = NULL;
1971                 }
1972         }
1973
1974         /* Resolve policies to use if we couldn't get them from
1975          * previous cache entry */
1976         if (xdst == NULL) {
1977                 num_pols = 1;
1978                 pols[0] = __xfrm_policy_lookup(net, fl, family,
1979                                                flow_to_policy_dir(dir));
1980                 err = xfrm_expand_policies(fl, family, pols,
1981                                            &num_pols, &num_xfrms);
1982                 if (err < 0)
1983                         goto inc_error;
1984                 if (num_pols == 0)
1985                         return NULL;
1986                 if (num_xfrms <= 0)
1987                         goto make_dummy_bundle;
1988         }
1989
1990         new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, dst_orig);
1991         if (IS_ERR(new_xdst)) {
1992                 err = PTR_ERR(new_xdst);
1993                 if (err != -EAGAIN)
1994                         goto error;
1995                 if (oldflo == NULL)
1996                         goto make_dummy_bundle;
1997                 dst_hold(&xdst->u.dst);
1998                 return oldflo;
1999         } else if (new_xdst == NULL) {
2000                 num_xfrms = 0;
2001                 if (oldflo == NULL)
2002                         goto make_dummy_bundle;
2003                 xdst->num_xfrms = 0;
2004                 dst_hold(&xdst->u.dst);
2005                 return oldflo;
2006         }
2007
2008         /* Kill the previous bundle */
2009         if (xdst) {
2010                 /* The policies were stolen for newly generated bundle */
2011                 xdst->num_pols = 0;
2012                 dst_free(&xdst->u.dst);
2013         }
2014
2015         /* Flow cache does not have reference, it dst_free()'s,
2016          * but we do need to return one reference for original caller */
2017         dst_hold(&new_xdst->u.dst);
2018         return &new_xdst->flo;
2019
2020 make_dummy_bundle:
2021         /* We found policies, but there's no bundles to instantiate:
2022          * either because the policy blocks, has no transformations or
2023          * we could not build template (no xfrm_states).*/
2024         xdst = xfrm_create_dummy_bundle(net, dst_orig, fl, num_xfrms, family);
2025         if (IS_ERR(xdst)) {
2026                 xfrm_pols_put(pols, num_pols);
2027                 return ERR_CAST(xdst);
2028         }
2029         xdst->num_pols = num_pols;
2030         xdst->num_xfrms = num_xfrms;
2031         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy*) * num_pols);
2032
2033         dst_hold(&xdst->u.dst);
2034         return &xdst->flo;
2035
2036 inc_error:
2037         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2038 error:
2039         if (xdst != NULL)
2040                 dst_free(&xdst->u.dst);
2041         else
2042                 xfrm_pols_put(pols, num_pols);
2043         return ERR_PTR(err);
2044 }
2045
2046 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2047                                         struct dst_entry *dst_orig)
2048 {
2049         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2050         struct dst_entry *ret;
2051
2052         if (!afinfo) {
2053                 dst_release(dst_orig);
2054                 return ERR_PTR(-EINVAL);
2055         } else {
2056                 ret = afinfo->blackhole_route(net, dst_orig);
2057         }
2058         xfrm_policy_put_afinfo(afinfo);
2059
2060         return ret;
2061 }
2062
2063 /* Main function: finds/creates a bundle for given flow.
2064  *
2065  * At the moment we eat a raw IP route. Mostly to speed up lookups
2066  * on interfaces with disabled IPsec.
2067  */
2068 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
2069                               const struct flowi *fl,
2070                               struct sock *sk, int flags)
2071 {
2072         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2073         struct flow_cache_object *flo;
2074         struct xfrm_dst *xdst;
2075         struct dst_entry *dst, *route;
2076         u16 family = dst_orig->ops->family;
2077         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
2078         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
2079
2080 restart:
2081         dst = NULL;
2082         xdst = NULL;
2083         route = NULL;
2084
2085         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
2086                 num_pols = 1;
2087                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
2088                 err = xfrm_expand_policies(fl, family, pols,
2089                                            &num_pols, &num_xfrms);
2090                 if (err < 0)
2091                         goto dropdst;
2092
2093                 if (num_pols) {
2094                         if (num_xfrms <= 0) {
2095                                 drop_pols = num_pols;
2096                                 goto no_transform;
2097                         }
2098
2099                         xdst = xfrm_resolve_and_create_bundle(
2100                                         pols, num_pols, fl,
2101                                         family, dst_orig);
2102                         if (IS_ERR(xdst)) {
2103                                 xfrm_pols_put(pols, num_pols);
2104                                 err = PTR_ERR(xdst);
2105                                 goto dropdst;
2106                         } else if (xdst == NULL) {
2107                                 num_xfrms = 0;
2108                                 drop_pols = num_pols;
2109                                 goto no_transform;
2110                         }
2111
2112                         dst_hold(&xdst->u.dst);
2113
2114                         spin_lock_bh(&net->xfrm.xfrm_policy_sk_bundle_lock);
2115                         xdst->u.dst.next = xfrm_policy_sk_bundles;
2116                         xfrm_policy_sk_bundles = &xdst->u.dst;
2117                         spin_unlock_bh(&net->xfrm.xfrm_policy_sk_bundle_lock);
2118
2119                         route = xdst->route;
2120                 }
2121         }
2122
2123         if (xdst == NULL) {
2124                 /* To accelerate a bit...  */
2125                 if ((dst_orig->flags & DST_NOXFRM) ||
2126                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
2127                         goto nopol;
2128
2129                 flo = flow_cache_lookup(net, fl, family, dir,
2130                                         xfrm_bundle_lookup, dst_orig);
2131                 if (flo == NULL)
2132                         goto nopol;
2133                 if (IS_ERR(flo)) {
2134                         err = PTR_ERR(flo);
2135                         goto dropdst;
2136                 }
2137                 xdst = container_of(flo, struct xfrm_dst, flo);
2138
2139                 num_pols = xdst->num_pols;
2140                 num_xfrms = xdst->num_xfrms;
2141                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy*) * num_pols);
2142                 route = xdst->route;
2143         }
2144
2145         dst = &xdst->u.dst;
2146         if (route == NULL && num_xfrms > 0) {
2147                 /* The only case when xfrm_bundle_lookup() returns a
2148                  * bundle with null route, is when the template could
2149                  * not be resolved. It means policies are there, but
2150                  * bundle could not be created, since we don't yet
2151                  * have the xfrm_state's. We need to wait for KM to
2152                  * negotiate new SA's or bail out with error.*/
2153                 if (net->xfrm.sysctl_larval_drop) {
2154                         dst_release(dst);
2155                         xfrm_pols_put(pols, drop_pols);
2156                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2157
2158                         return make_blackhole(net, family, dst_orig);
2159                 }
2160                 if (fl->flowi_flags & FLOWI_FLAG_CAN_SLEEP) {
2161                         DECLARE_WAITQUEUE(wait, current);
2162
2163                         add_wait_queue(&net->xfrm.km_waitq, &wait);
2164                         set_current_state(TASK_INTERRUPTIBLE);
2165                         schedule();
2166                         set_current_state(TASK_RUNNING);
2167                         remove_wait_queue(&net->xfrm.km_waitq, &wait);
2168
2169                         if (!signal_pending(current)) {
2170                                 dst_release(dst);
2171                                 goto restart;
2172                         }
2173
2174                         err = -ERESTART;
2175                 } else
2176                         err = -EAGAIN;
2177
2178                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
2179                 goto error;
2180         }
2181
2182 no_transform:
2183         if (num_pols == 0)
2184                 goto nopol;
2185
2186         if ((flags & XFRM_LOOKUP_ICMP) &&
2187             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
2188                 err = -ENOENT;
2189                 goto error;
2190         }
2191
2192         for (i = 0; i < num_pols; i++)
2193                 pols[i]->curlft.use_time = get_seconds();
2194
2195         if (num_xfrms < 0) {
2196                 /* Prohibit the flow */
2197                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
2198                 err = -EPERM;
2199                 goto error;
2200         } else if (num_xfrms > 0) {
2201                 /* Flow transformed */
2202                 dst_release(dst_orig);
2203         } else {
2204                 /* Flow passes untransformed */
2205                 dst_release(dst);
2206                 dst = dst_orig;
2207         }
2208 ok:
2209         xfrm_pols_put(pols, drop_pols);
2210         if (dst && dst->xfrm &&
2211             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
2212                 dst->flags |= DST_XFRM_TUNNEL;
2213         return dst;
2214
2215 nopol:
2216         if (!(flags & XFRM_LOOKUP_ICMP)) {
2217                 dst = dst_orig;
2218                 goto ok;
2219         }
2220         err = -ENOENT;
2221 error:
2222         dst_release(dst);
2223 dropdst:
2224         dst_release(dst_orig);
2225         xfrm_pols_put(pols, drop_pols);
2226         return ERR_PTR(err);
2227 }
2228 EXPORT_SYMBOL(xfrm_lookup);
2229
2230 static inline int
2231 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
2232 {
2233         struct xfrm_state *x;
2234
2235         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
2236                 return 0;
2237         x = skb->sp->xvec[idx];
2238         if (!x->type->reject)
2239                 return 0;
2240         return x->type->reject(x, skb, fl);
2241 }
2242
2243 /* When skb is transformed back to its "native" form, we have to
2244  * check policy restrictions. At the moment we make this in maximally
2245  * stupid way. Shame on me. :-) Of course, connected sockets must
2246  * have policy cached at them.
2247  */
2248
2249 static inline int
2250 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
2251               unsigned short family)
2252 {
2253         if (xfrm_state_kern(x))
2254                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
2255         return  x->id.proto == tmpl->id.proto &&
2256                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
2257                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
2258                 x->props.mode == tmpl->mode &&
2259                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
2260                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
2261                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
2262                   xfrm_state_addr_cmp(tmpl, x, family));
2263 }
2264
2265 /*
2266  * 0 or more than 0 is returned when validation is succeeded (either bypass
2267  * because of optional transport mode, or next index of the mathced secpath
2268  * state with the template.
2269  * -1 is returned when no matching template is found.
2270  * Otherwise "-2 - errored_index" is returned.
2271  */
2272 static inline int
2273 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
2274                unsigned short family)
2275 {
2276         int idx = start;
2277
2278         if (tmpl->optional) {
2279                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
2280                         return start;
2281         } else
2282                 start = -1;
2283         for (; idx < sp->len; idx++) {
2284                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
2285                         return ++idx;
2286                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
2287                         if (start == -1)
2288                                 start = -2-idx;
2289                         break;
2290                 }
2291         }
2292         return start;
2293 }
2294
2295 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
2296                           unsigned int family, int reverse)
2297 {
2298         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2299         int err;
2300
2301         if (unlikely(afinfo == NULL))
2302                 return -EAFNOSUPPORT;
2303
2304         afinfo->decode_session(skb, fl, reverse);
2305         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
2306         xfrm_policy_put_afinfo(afinfo);
2307         return err;
2308 }
2309 EXPORT_SYMBOL(__xfrm_decode_session);
2310
2311 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
2312 {
2313         for (; k < sp->len; k++) {
2314                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
2315                         *idxp = k;
2316                         return 1;
2317                 }
2318         }
2319
2320         return 0;
2321 }
2322
2323 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
2324                         unsigned short family)
2325 {
2326         struct net *net = dev_net(skb->dev);
2327         struct xfrm_policy *pol;
2328         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2329         int npols = 0;
2330         int xfrm_nr;
2331         int pi;
2332         int reverse;
2333         struct flowi fl;
2334         u8 fl_dir;
2335         int xerr_idx = -1;
2336
2337         reverse = dir & ~XFRM_POLICY_MASK;
2338         dir &= XFRM_POLICY_MASK;
2339         fl_dir = policy_to_flow_dir(dir);
2340
2341         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
2342                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
2343                 return 0;
2344         }
2345
2346         nf_nat_decode_session(skb, &fl, family);
2347
2348         /* First, check used SA against their selectors. */
2349         if (skb->sp) {
2350                 int i;
2351
2352                 for (i=skb->sp->len-1; i>=0; i--) {
2353                         struct xfrm_state *x = skb->sp->xvec[i];
2354                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
2355                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
2356                                 return 0;
2357                         }
2358                 }
2359         }
2360
2361         pol = NULL;
2362         if (sk && sk->sk_policy[dir]) {
2363                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
2364                 if (IS_ERR(pol)) {
2365                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2366                         return 0;
2367                 }
2368         }
2369
2370         if (!pol) {
2371                 struct flow_cache_object *flo;
2372
2373                 flo = flow_cache_lookup(net, &fl, family, fl_dir,
2374                                         xfrm_policy_lookup, NULL);
2375                 if (IS_ERR_OR_NULL(flo))
2376                         pol = ERR_CAST(flo);
2377                 else
2378                         pol = container_of(flo, struct xfrm_policy, flo);
2379         }
2380
2381         if (IS_ERR(pol)) {
2382                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2383                 return 0;
2384         }
2385
2386         if (!pol) {
2387                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
2388                         xfrm_secpath_reject(xerr_idx, skb, &fl);
2389                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
2390                         return 0;
2391                 }
2392                 return 1;
2393         }
2394
2395         pol->curlft.use_time = get_seconds();
2396
2397         pols[0] = pol;
2398         npols ++;
2399 #ifdef CONFIG_XFRM_SUB_POLICY
2400         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2401                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
2402                                                     &fl, family,
2403                                                     XFRM_POLICY_IN);
2404                 if (pols[1]) {
2405                         if (IS_ERR(pols[1])) {
2406                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
2407                                 return 0;
2408                         }
2409                         pols[1]->curlft.use_time = get_seconds();
2410                         npols ++;
2411                 }
2412         }
2413 #endif
2414
2415         if (pol->action == XFRM_POLICY_ALLOW) {
2416                 struct sec_path *sp;
2417                 static struct sec_path dummy;
2418                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
2419                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
2420                 struct xfrm_tmpl **tpp = tp;
2421                 int ti = 0;
2422                 int i, k;
2423
2424                 if ((sp = skb->sp) == NULL)
2425                         sp = &dummy;
2426
2427                 for (pi = 0; pi < npols; pi++) {
2428                         if (pols[pi] != pol &&
2429                             pols[pi]->action != XFRM_POLICY_ALLOW) {
2430                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2431                                 goto reject;
2432                         }
2433                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
2434                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
2435                                 goto reject_error;
2436                         }
2437                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
2438                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
2439                 }
2440                 xfrm_nr = ti;
2441                 if (npols > 1) {
2442                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
2443                         tpp = stp;
2444                 }
2445
2446                 /* For each tunnel xfrm, find the first matching tmpl.
2447                  * For each tmpl before that, find corresponding xfrm.
2448                  * Order is _important_. Later we will implement
2449                  * some barriers, but at the moment barriers
2450                  * are implied between each two transformations.
2451                  */
2452                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
2453                         k = xfrm_policy_ok(tpp[i], sp, k, family);
2454                         if (k < 0) {
2455                                 if (k < -1)
2456                                         /* "-2 - errored_index" returned */
2457                                         xerr_idx = -(2+k);
2458                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2459                                 goto reject;
2460                         }
2461                 }
2462
2463                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
2464                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
2465                         goto reject;
2466                 }
2467
2468                 xfrm_pols_put(pols, npols);
2469                 return 1;
2470         }
2471         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
2472
2473 reject:
2474         xfrm_secpath_reject(xerr_idx, skb, &fl);
2475 reject_error:
2476         xfrm_pols_put(pols, npols);
2477         return 0;
2478 }
2479 EXPORT_SYMBOL(__xfrm_policy_check);
2480
2481 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
2482 {
2483         struct net *net = dev_net(skb->dev);
2484         struct flowi fl;
2485         struct dst_entry *dst;
2486         int res = 1;
2487
2488         if (xfrm_decode_session(skb, &fl, family) < 0) {
2489                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
2490                 return 0;
2491         }
2492
2493         skb_dst_force(skb);
2494
2495         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, 0);
2496         if (IS_ERR(dst)) {
2497                 res = 0;
2498                 dst = NULL;
2499         }
2500         skb_dst_set(skb, dst);
2501         return res;
2502 }
2503 EXPORT_SYMBOL(__xfrm_route_forward);
2504
2505 /* Optimize later using cookies and generation ids. */
2506
2507 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
2508 {
2509         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
2510          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
2511          * get validated by dst_ops->check on every use.  We do this
2512          * because when a normal route referenced by an XFRM dst is
2513          * obsoleted we do not go looking around for all parent
2514          * referencing XFRM dsts so that we can invalidate them.  It
2515          * is just too much work.  Instead we make the checks here on
2516          * every use.  For example:
2517          *
2518          *      XFRM dst A --> IPv4 dst X
2519          *
2520          * X is the "xdst->route" of A (X is also the "dst->path" of A
2521          * in this example).  If X is marked obsolete, "A" will not
2522          * notice.  That's what we are validating here via the
2523          * stale_bundle() check.
2524          *
2525          * When a policy's bundle is pruned, we dst_free() the XFRM
2526          * dst which causes it's ->obsolete field to be set to
2527          * DST_OBSOLETE_DEAD.  If an XFRM dst has been pruned like
2528          * this, we want to force a new route lookup.
2529          */
2530         if (dst->obsolete < 0 && !stale_bundle(dst))
2531                 return dst;
2532
2533         return NULL;
2534 }
2535
2536 static int stale_bundle(struct dst_entry *dst)
2537 {
2538         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
2539 }
2540
2541 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
2542 {
2543         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
2544                 dst->dev = dev_net(dev)->loopback_dev;
2545                 dev_hold(dst->dev);
2546                 dev_put(dev);
2547         }
2548 }
2549 EXPORT_SYMBOL(xfrm_dst_ifdown);
2550
2551 static void xfrm_link_failure(struct sk_buff *skb)
2552 {
2553         /* Impossible. Such dst must be popped before reaches point of failure. */
2554 }
2555
2556 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
2557 {
2558         if (dst) {
2559                 if (dst->obsolete) {
2560                         dst_release(dst);
2561                         dst = NULL;
2562                 }
2563         }
2564         return dst;
2565 }
2566
2567 static void __xfrm_garbage_collect(struct net *net)
2568 {
2569         struct dst_entry *head, *next;
2570
2571         spin_lock_bh(&net->xfrm.xfrm_policy_sk_bundle_lock);
2572         head = xfrm_policy_sk_bundles;
2573         xfrm_policy_sk_bundles = NULL;
2574         spin_unlock_bh(&net->xfrm.xfrm_policy_sk_bundle_lock);
2575
2576         while (head) {
2577                 next = head->next;
2578                 dst_free(head);
2579                 head = next;
2580         }
2581 }
2582
2583 void xfrm_garbage_collect(struct net *net)
2584 {
2585         flow_cache_flush();
2586         __xfrm_garbage_collect(net);
2587 }
2588 EXPORT_SYMBOL(xfrm_garbage_collect);
2589
2590 static void xfrm_garbage_collect_deferred(struct net *net)
2591 {
2592         flow_cache_flush_deferred();
2593         __xfrm_garbage_collect(net);
2594 }
2595
2596 static void xfrm_init_pmtu(struct dst_entry *dst)
2597 {
2598         do {
2599                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2600                 u32 pmtu, route_mtu_cached;
2601
2602                 pmtu = dst_mtu(dst->child);
2603                 xdst->child_mtu_cached = pmtu;
2604
2605                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
2606
2607                 route_mtu_cached = dst_mtu(xdst->route);
2608                 xdst->route_mtu_cached = route_mtu_cached;
2609
2610                 if (pmtu > route_mtu_cached)
2611                         pmtu = route_mtu_cached;
2612
2613                 dst_metric_set(dst, RTAX_MTU, pmtu);
2614         } while ((dst = dst->next));
2615 }
2616
2617 /* Check that the bundle accepts the flow and its components are
2618  * still valid.
2619  */
2620
2621 static int xfrm_bundle_ok(struct xfrm_dst *first)
2622 {
2623         struct dst_entry *dst = &first->u.dst;
2624         struct xfrm_dst *last;
2625         u32 mtu;
2626
2627         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
2628             (dst->dev && !netif_running(dst->dev)))
2629                 return 0;
2630
2631         if (dst->flags & DST_XFRM_QUEUE)
2632                 return 1;
2633
2634         last = NULL;
2635
2636         do {
2637                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
2638
2639                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
2640                         return 0;
2641                 if (xdst->xfrm_genid != dst->xfrm->genid)
2642                         return 0;
2643                 if (xdst->num_pols > 0 &&
2644                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
2645                         return 0;
2646
2647                 mtu = dst_mtu(dst->child);
2648                 if (xdst->child_mtu_cached != mtu) {
2649                         last = xdst;
2650                         xdst->child_mtu_cached = mtu;
2651                 }
2652
2653                 if (!dst_check(xdst->route, xdst->route_cookie))
2654                         return 0;
2655                 mtu = dst_mtu(xdst->route);
2656                 if (xdst->route_mtu_cached != mtu) {
2657                         last = xdst;
2658                         xdst->route_mtu_cached = mtu;
2659                 }
2660
2661                 dst = dst->child;
2662         } while (dst->xfrm);
2663
2664         if (likely(!last))
2665                 return 1;
2666
2667         mtu = last->child_mtu_cached;
2668         for (;;) {
2669                 dst = &last->u.dst;
2670
2671                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
2672                 if (mtu > last->route_mtu_cached)
2673                         mtu = last->route_mtu_cached;
2674                 dst_metric_set(dst, RTAX_MTU, mtu);
2675
2676                 if (last == first)
2677                         break;
2678
2679                 last = (struct xfrm_dst *)last->u.dst.next;
2680                 last->child_mtu_cached = mtu;
2681         }
2682
2683         return 1;
2684 }
2685
2686 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
2687 {
2688         return dst_metric_advmss(dst->path);
2689 }
2690
2691 static unsigned int xfrm_mtu(const struct dst_entry *dst)
2692 {
2693         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2694
2695         return mtu ? : dst_mtu(dst->path);
2696 }
2697
2698 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
2699                                            struct sk_buff *skb,
2700                                            const void *daddr)
2701 {
2702         return dst->path->ops->neigh_lookup(dst, skb, daddr);
2703 }
2704
2705 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
2706 {
2707         struct net *net;
2708         int err = 0;
2709         if (unlikely(afinfo == NULL))
2710                 return -EINVAL;
2711         if (unlikely(afinfo->family >= NPROTO))
2712                 return -EAFNOSUPPORT;
2713         spin_lock(&xfrm_policy_afinfo_lock);
2714         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
2715                 err = -ENOBUFS;
2716         else {
2717                 struct dst_ops *dst_ops = afinfo->dst_ops;
2718                 if (likely(dst_ops->kmem_cachep == NULL))
2719                         dst_ops->kmem_cachep = xfrm_dst_cache;
2720                 if (likely(dst_ops->check == NULL))
2721                         dst_ops->check = xfrm_dst_check;
2722                 if (likely(dst_ops->default_advmss == NULL))
2723                         dst_ops->default_advmss = xfrm_default_advmss;
2724                 if (likely(dst_ops->mtu == NULL))
2725                         dst_ops->mtu = xfrm_mtu;
2726                 if (likely(dst_ops->negative_advice == NULL))
2727                         dst_ops->negative_advice = xfrm_negative_advice;
2728                 if (likely(dst_ops->link_failure == NULL))
2729                         dst_ops->link_failure = xfrm_link_failure;
2730                 if (likely(dst_ops->neigh_lookup == NULL))
2731                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
2732                 if (likely(afinfo->garbage_collect == NULL))
2733                         afinfo->garbage_collect = xfrm_garbage_collect_deferred;
2734                 rcu_assign_pointer(xfrm_policy_afinfo[afinfo->family], afinfo);
2735         }
2736         spin_unlock(&xfrm_policy_afinfo_lock);
2737
2738         rtnl_lock();
2739         for_each_net(net) {
2740                 struct dst_ops *xfrm_dst_ops;
2741
2742                 switch (afinfo->family) {
2743                 case AF_INET:
2744                         xfrm_dst_ops = &net->xfrm.xfrm4_dst_ops;
2745                         break;
2746 #if IS_ENABLED(CONFIG_IPV6)
2747                 case AF_INET6:
2748                         xfrm_dst_ops = &net->xfrm.xfrm6_dst_ops;
2749                         break;
2750 #endif
2751                 default:
2752                         BUG();
2753                 }
2754                 *xfrm_dst_ops = *afinfo->dst_ops;
2755         }
2756         rtnl_unlock();
2757
2758         return err;
2759 }
2760 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2761
2762 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2763 {
2764         int err = 0;
2765         if (unlikely(afinfo == NULL))
2766                 return -EINVAL;
2767         if (unlikely(afinfo->family >= NPROTO))
2768                 return -EAFNOSUPPORT;
2769         spin_lock(&xfrm_policy_afinfo_lock);
2770         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2771                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2772                         err = -EINVAL;
2773                 else
2774                         RCU_INIT_POINTER(xfrm_policy_afinfo[afinfo->family],
2775                                          NULL);
2776         }
2777         spin_unlock(&xfrm_policy_afinfo_lock);
2778         if (!err) {
2779                 struct dst_ops *dst_ops = afinfo->dst_ops;
2780
2781                 synchronize_rcu();
2782
2783                 dst_ops->kmem_cachep = NULL;
2784                 dst_ops->check = NULL;
2785                 dst_ops->negative_advice = NULL;
2786                 dst_ops->link_failure = NULL;
2787                 afinfo->garbage_collect = NULL;
2788         }
2789         return err;
2790 }
2791 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2792
2793 static void __net_init xfrm_dst_ops_init(struct net *net)
2794 {
2795         struct xfrm_policy_afinfo *afinfo;
2796
2797         rcu_read_lock();
2798         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET]);
2799         if (afinfo)
2800                 net->xfrm.xfrm4_dst_ops = *afinfo->dst_ops;
2801 #if IS_ENABLED(CONFIG_IPV6)
2802         afinfo = rcu_dereference(xfrm_policy_afinfo[AF_INET6]);
2803         if (afinfo)
2804                 net->xfrm.xfrm6_dst_ops = *afinfo->dst_ops;
2805 #endif
2806         rcu_read_unlock();
2807 }
2808
2809 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2810 {
2811         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2812
2813         switch (event) {
2814         case NETDEV_DOWN:
2815                 xfrm_garbage_collect(dev_net(dev));
2816         }
2817         return NOTIFY_DONE;
2818 }
2819
2820 static struct notifier_block xfrm_dev_notifier = {
2821         .notifier_call  = xfrm_dev_event,
2822 };
2823
2824 #ifdef CONFIG_XFRM_STATISTICS
2825 static int __net_init xfrm_statistics_init(struct net *net)
2826 {
2827         int rv;
2828
2829         if (snmp_mib_init((void __percpu **)net->mib.xfrm_statistics,
2830                           sizeof(struct linux_xfrm_mib),
2831                           __alignof__(struct linux_xfrm_mib)) < 0)
2832                 return -ENOMEM;
2833         rv = xfrm_proc_init(net);
2834         if (rv < 0)
2835                 snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2836         return rv;
2837 }
2838
2839 static void xfrm_statistics_fini(struct net *net)
2840 {
2841         xfrm_proc_fini(net);
2842         snmp_mib_free((void __percpu **)net->mib.xfrm_statistics);
2843 }
2844 #else
2845 static int __net_init xfrm_statistics_init(struct net *net)
2846 {
2847         return 0;
2848 }
2849
2850 static void xfrm_statistics_fini(struct net *net)
2851 {
2852 }
2853 #endif
2854
2855 static int __net_init xfrm_policy_init(struct net *net)
2856 {
2857         unsigned int hmask, sz;
2858         int dir;
2859
2860         if (net_eq(net, &init_net))
2861                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2862                                            sizeof(struct xfrm_dst),
2863                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2864                                            NULL);
2865
2866         hmask = 8 - 1;
2867         sz = (hmask+1) * sizeof(struct hlist_head);
2868
2869         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
2870         if (!net->xfrm.policy_byidx)
2871                 goto out_byidx;
2872         net->xfrm.policy_idx_hmask = hmask;
2873
2874         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2875                 struct xfrm_policy_hash *htab;
2876
2877                 net->xfrm.policy_count[dir] = 0;
2878                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
2879
2880                 htab = &net->xfrm.policy_bydst[dir];
2881                 htab->table = xfrm_hash_alloc(sz);
2882                 if (!htab->table)
2883                         goto out_bydst;
2884                 htab->hmask = hmask;
2885         }
2886
2887         INIT_LIST_HEAD(&net->xfrm.policy_all);
2888         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
2889         if (net_eq(net, &init_net))
2890                 register_netdevice_notifier(&xfrm_dev_notifier);
2891         return 0;
2892
2893 out_bydst:
2894         for (dir--; dir >= 0; dir--) {
2895                 struct xfrm_policy_hash *htab;
2896
2897                 htab = &net->xfrm.policy_bydst[dir];
2898                 xfrm_hash_free(htab->table, sz);
2899         }
2900         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2901 out_byidx:
2902         return -ENOMEM;
2903 }
2904
2905 static void xfrm_policy_fini(struct net *net)
2906 {
2907         struct xfrm_audit audit_info;
2908         unsigned int sz;
2909         int dir;
2910
2911         flush_work(&net->xfrm.policy_hash_work);
2912 #ifdef CONFIG_XFRM_SUB_POLICY
2913         audit_info.loginuid = INVALID_UID;
2914         audit_info.sessionid = -1;
2915         audit_info.secid = 0;
2916         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
2917 #endif
2918         audit_info.loginuid = INVALID_UID;
2919         audit_info.sessionid = -1;
2920         audit_info.secid = 0;
2921         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
2922
2923         WARN_ON(!list_empty(&net->xfrm.policy_all));
2924
2925         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2926                 struct xfrm_policy_hash *htab;
2927
2928                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
2929
2930                 htab = &net->xfrm.policy_bydst[dir];
2931                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
2932                 WARN_ON(!hlist_empty(htab->table));
2933                 xfrm_hash_free(htab->table, sz);
2934         }
2935
2936         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
2937         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
2938         xfrm_hash_free(net->xfrm.policy_byidx, sz);
2939 }
2940
2941 static int __net_init xfrm_net_init(struct net *net)
2942 {
2943         int rv;
2944
2945         rv = xfrm_statistics_init(net);
2946         if (rv < 0)
2947                 goto out_statistics;
2948         rv = xfrm_state_init(net);
2949         if (rv < 0)
2950                 goto out_state;
2951         rv = xfrm_policy_init(net);
2952         if (rv < 0)
2953                 goto out_policy;
2954         xfrm_dst_ops_init(net);
2955         rv = xfrm_sysctl_init(net);
2956         if (rv < 0)
2957                 goto out_sysctl;
2958
2959         /* Initialize the per-net locks here */
2960         spin_lock_init(&net->xfrm.xfrm_state_lock);
2961         rwlock_init(&net->xfrm.xfrm_policy_lock);
2962         spin_lock_init(&net->xfrm.xfrm_policy_sk_bundle_lock);
2963         mutex_init(&net->xfrm.xfrm_cfg_mutex);
2964
2965         return 0;
2966
2967 out_sysctl:
2968         xfrm_policy_fini(net);
2969 out_policy:
2970         xfrm_state_fini(net);
2971 out_state:
2972         xfrm_statistics_fini(net);
2973 out_statistics:
2974         return rv;
2975 }
2976
2977 static void __net_exit xfrm_net_exit(struct net *net)
2978 {
2979         xfrm_sysctl_fini(net);
2980         xfrm_policy_fini(net);
2981         xfrm_state_fini(net);
2982         xfrm_statistics_fini(net);
2983 }
2984
2985 static struct pernet_operations __net_initdata xfrm_net_ops = {
2986         .init = xfrm_net_init,
2987         .exit = xfrm_net_exit,
2988 };
2989
2990 void __init xfrm_init(void)
2991 {
2992         register_pernet_subsys(&xfrm_net_ops);
2993         xfrm_input_init();
2994 }
2995
2996 #ifdef CONFIG_AUDITSYSCALL
2997 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
2998                                          struct audit_buffer *audit_buf)
2999 {
3000         struct xfrm_sec_ctx *ctx = xp->security;
3001         struct xfrm_selector *sel = &xp->selector;
3002
3003         if (ctx)
3004                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3005                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3006
3007         switch(sel->family) {
3008         case AF_INET:
3009                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3010                 if (sel->prefixlen_s != 32)
3011                         audit_log_format(audit_buf, " src_prefixlen=%d",
3012                                          sel->prefixlen_s);
3013                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3014                 if (sel->prefixlen_d != 32)
3015                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3016                                          sel->prefixlen_d);
3017                 break;
3018         case AF_INET6:
3019                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3020                 if (sel->prefixlen_s != 128)
3021                         audit_log_format(audit_buf, " src_prefixlen=%d",
3022                                          sel->prefixlen_s);
3023                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3024                 if (sel->prefixlen_d != 128)
3025                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3026                                          sel->prefixlen_d);
3027                 break;
3028         }
3029 }
3030
3031 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result,
3032                            kuid_t auid, u32 sessionid, u32 secid)
3033 {
3034         struct audit_buffer *audit_buf;
3035
3036         audit_buf = xfrm_audit_start("SPD-add");
3037         if (audit_buf == NULL)
3038                 return;
3039         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3040         audit_log_format(audit_buf, " res=%u", result);
3041         xfrm_audit_common_policyinfo(xp, audit_buf);
3042         audit_log_end(audit_buf);
3043 }
3044 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
3045
3046 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
3047                               kuid_t auid, u32 sessionid, u32 secid)
3048 {
3049         struct audit_buffer *audit_buf;
3050
3051         audit_buf = xfrm_audit_start("SPD-delete");
3052         if (audit_buf == NULL)
3053                 return;
3054         xfrm_audit_helper_usrinfo(auid, sessionid, secid, audit_buf);
3055         audit_log_format(audit_buf, " res=%u", result);
3056         xfrm_audit_common_policyinfo(xp, audit_buf);
3057         audit_log_end(audit_buf);
3058 }
3059 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
3060 #endif
3061
3062 #ifdef CONFIG_XFRM_MIGRATE
3063 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
3064                                         const struct xfrm_selector *sel_tgt)
3065 {
3066         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
3067                 if (sel_tgt->family == sel_cmp->family &&
3068                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
3069                                     sel_cmp->family) &&
3070                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
3071                                     sel_cmp->family) &&
3072                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
3073                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
3074                         return true;
3075                 }
3076         } else {
3077                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
3078                         return true;
3079                 }
3080         }
3081         return false;
3082 }
3083
3084 static struct xfrm_policy * xfrm_migrate_policy_find(const struct xfrm_selector *sel,
3085                                                      u8 dir, u8 type, struct net *net)
3086 {
3087         struct xfrm_policy *pol, *ret = NULL;
3088         struct hlist_head *chain;
3089         u32 priority = ~0U;
3090
3091         read_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME*/
3092         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
3093         hlist_for_each_entry(pol, chain, bydst) {
3094                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3095                     pol->type == type) {
3096                         ret = pol;
3097                         priority = ret->priority;
3098                         break;
3099                 }
3100         }
3101         chain = &net->xfrm.policy_inexact[dir];
3102         hlist_for_each_entry(pol, chain, bydst) {
3103                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
3104                     pol->type == type &&
3105                     pol->priority < priority) {
3106                         ret = pol;
3107                         break;
3108                 }
3109         }
3110
3111         if (ret)
3112                 xfrm_pol_hold(ret);
3113
3114         read_unlock_bh(&net->xfrm.xfrm_policy_lock);
3115
3116         return ret;
3117 }
3118
3119 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
3120 {
3121         int match = 0;
3122
3123         if (t->mode == m->mode && t->id.proto == m->proto &&
3124             (m->reqid == 0 || t->reqid == m->reqid)) {
3125                 switch (t->mode) {
3126                 case XFRM_MODE_TUNNEL:
3127                 case XFRM_MODE_BEET:
3128                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
3129                                             m->old_family) &&
3130                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
3131                                             m->old_family)) {
3132                                 match = 1;
3133                         }
3134                         break;
3135                 case XFRM_MODE_TRANSPORT:
3136                         /* in case of transport mode, template does not store
3137                            any IP addresses, hence we just compare mode and
3138                            protocol */
3139                         match = 1;
3140                         break;
3141                 default:
3142                         break;
3143                 }
3144         }
3145         return match;
3146 }
3147
3148 /* update endpoint address(es) of template(s) */
3149 static int xfrm_policy_migrate(struct xfrm_policy *pol,
3150                                struct xfrm_migrate *m, int num_migrate)
3151 {
3152         struct xfrm_migrate *mp;
3153         int i, j, n = 0;
3154
3155         write_lock_bh(&pol->lock);
3156         if (unlikely(pol->walk.dead)) {
3157                 /* target policy has been deleted */
3158                 write_unlock_bh(&pol->lock);
3159                 return -ENOENT;
3160         }
3161
3162         for (i = 0; i < pol->xfrm_nr; i++) {
3163                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
3164                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
3165                                 continue;
3166                         n++;
3167                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
3168                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
3169                                 continue;
3170                         /* update endpoints */
3171                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
3172                                sizeof(pol->xfrm_vec[i].id.daddr));
3173                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
3174                                sizeof(pol->xfrm_vec[i].saddr));
3175                         pol->xfrm_vec[i].encap_family = mp->new_family;
3176                         /* flush bundles */
3177                         atomic_inc(&pol->genid);
3178                 }
3179         }
3180
3181         write_unlock_bh(&pol->lock);
3182
3183         if (!n)
3184                 return -ENODATA;
3185
3186         return 0;
3187 }
3188
3189 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
3190 {
3191         int i, j;
3192
3193         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
3194                 return -EINVAL;
3195
3196         for (i = 0; i < num_migrate; i++) {
3197                 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr,
3198                                     m[i].old_family) &&
3199                     xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr,
3200                                     m[i].old_family))
3201                         return -EINVAL;
3202                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
3203                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
3204                         return -EINVAL;
3205
3206                 /* check if there is any duplicated entry */
3207                 for (j = i + 1; j < num_migrate; j++) {
3208                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
3209                                     sizeof(m[i].old_daddr)) &&
3210                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
3211                                     sizeof(m[i].old_saddr)) &&
3212                             m[i].proto == m[j].proto &&
3213                             m[i].mode == m[j].mode &&
3214                             m[i].reqid == m[j].reqid &&
3215                             m[i].old_family == m[j].old_family)
3216                                 return -EINVAL;
3217                 }
3218         }
3219
3220         return 0;
3221 }
3222
3223 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
3224                  struct xfrm_migrate *m, int num_migrate,
3225                  struct xfrm_kmaddress *k, struct net *net)
3226 {
3227         int i, err, nx_cur = 0, nx_new = 0;
3228         struct xfrm_policy *pol = NULL;
3229         struct xfrm_state *x, *xc;
3230         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
3231         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
3232         struct xfrm_migrate *mp;
3233
3234         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
3235                 goto out;
3236
3237         /* Stage 1 - find policy */
3238         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
3239                 err = -ENOENT;
3240                 goto out;
3241         }
3242
3243         /* Stage 2 - find and update state(s) */
3244         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
3245                 if ((x = xfrm_migrate_state_find(mp, net))) {
3246                         x_cur[nx_cur] = x;
3247                         nx_cur++;
3248                         if ((xc = xfrm_state_migrate(x, mp))) {
3249                                 x_new[nx_new] = xc;
3250                                 nx_new++;
3251                         } else {
3252                                 err = -ENODATA;
3253                                 goto restore_state;
3254                         }
3255                 }
3256         }
3257
3258         /* Stage 3 - update policy */
3259         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
3260                 goto restore_state;
3261
3262         /* Stage 4 - delete old state(s) */
3263         if (nx_cur) {
3264                 xfrm_states_put(x_cur, nx_cur);
3265                 xfrm_states_delete(x_cur, nx_cur);
3266         }
3267
3268         /* Stage 5 - announce */
3269         km_migrate(sel, dir, type, m, num_migrate, k);
3270
3271         xfrm_pol_put(pol);
3272
3273         return 0;
3274 out:
3275         return err;
3276
3277 restore_state:
3278         if (pol)
3279                 xfrm_pol_put(pol);
3280         if (nx_cur)
3281                 xfrm_states_put(x_cur, nx_cur);
3282         if (nx_new)
3283                 xfrm_states_delete(x_new, nx_new);
3284
3285         return err;
3286 }
3287 EXPORT_SYMBOL(xfrm_migrate);
3288 #endif