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/cpu.h>
  28 #include <linux/audit.h>
  29 #include <linux/rhashtable.h>
  30 #include <net/dst.h>
  31 #include <net/flow.h>
  32 #include <net/xfrm.h>
  33 #include <net/ip.h>
  34 #ifdef CONFIG_XFRM_STATISTICS
  35 #include <net/snmp.h>
  36 #endif
  37
  38 #include "xfrm_hash.h"
  39
  40 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  41 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  42 #define XFRM_MAX_QUEUE_LEN      100
  43
  44 struct xfrm_flo {
  45         struct dst_entry *dst_orig;
  46         u8 flags;
  47 };
  48
  49 /* prefixes smaller than this are stored in lists, not trees. */
  50 #define INEXACT_PREFIXLEN_IPV4  16
  51 #define INEXACT_PREFIXLEN_IPV6  48
  52
  53 struct xfrm_pol_inexact_node {
  54         struct rb_node node;
  55         union {
  56                 xfrm_address_t addr;
  57                 struct rcu_head rcu;
  58         };
  59         u8 prefixlen;
  60
  61         struct rb_root root;
  62
  63         /* the policies matching this node, can be empty list */
  64         struct hlist_head hhead;
  65 };
  66
  67 /* xfrm inexact policy search tree:
  68  * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
  69  *  |
  70  * +---- root_d: sorted by daddr:prefix
  71  * |                 |
  72  * |        xfrm_pol_inexact_node
  73  * |                 |
  74  * |                 +- root: sorted by saddr/prefix
  75  * |                 |              |
  76  * |                 |         xfrm_pol_inexact_node
  77  * |                 |              |
  78  * |                 |              + root: unused
  79  * |                 |              |
  80  * |                 |              + hhead: saddr:daddr policies
  81  * |                 |
  82  * |                 +- coarse policies and all any:daddr policies
  83  * |
  84  * +---- root_s: sorted by saddr:prefix
  85  * |                 |
  86  * |        xfrm_pol_inexact_node
  87  * |                 |
  88  * |                 + root: unused
  89  * |                 |
  90  * |                 + hhead: saddr:any policies
  91  * |
  92  * +---- coarse policies and all any:any policies
  93  *
  94  * Lookups return four candidate lists:
  95  * 1. any:any list from top-level xfrm_pol_inexact_bin
  96  * 2. any:daddr list from daddr tree
  97  * 3. saddr:daddr list from 2nd level daddr tree
  98  * 4. saddr:any list from saddr tree
  99  *
 100  * This result set then needs to be searched for the policy with
 101  * the lowest priority.  If two results have same prio, youngest one wins.
 102  */
 103
 104 struct xfrm_pol_inexact_key {
 105         possible_net_t net;
 106         u32 if_id;
 107         u16 family;
 108         u8 dir, type;
 109 };
 110
 111 struct xfrm_pol_inexact_bin {
 112         struct xfrm_pol_inexact_key k;
 113         struct rhash_head head;
 114         /* list containing '*:*' policies */
 115         struct hlist_head hhead;
 116
 117         seqcount_t count;
 118         /* tree sorted by daddr/prefix */
 119         struct rb_root root_d;
 120
 121         /* tree sorted by saddr/prefix */
 122         struct rb_root root_s;
 123
 124         /* slow path below */
 125         struct list_head inexact_bins;
 126         struct rcu_head rcu;
 127 };
 128
 129 enum xfrm_pol_inexact_candidate_type {
 130         XFRM_POL_CAND_BOTH,
 131         XFRM_POL_CAND_SADDR,
 132         XFRM_POL_CAND_DADDR,
 133         XFRM_POL_CAND_ANY,
 134
 135         XFRM_POL_CAND_MAX,
 136 };
 137
 138 struct xfrm_pol_inexact_candidates {
 139         struct hlist_head *res[XFRM_POL_CAND_MAX];
 140 };
 141
 142 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
 143 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
 144
 145 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
 146 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
 147                                                 __read_mostly;
 148
 149 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
 150 static __read_mostly seqcount_t xfrm_policy_hash_generation;
 151
 152 static struct rhashtable xfrm_policy_inexact_table;
 153 static const struct rhashtable_params xfrm_pol_inexact_params;
 154
 155 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
 156 static int stale_bundle(struct dst_entry *dst);
 157 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
 158 static void xfrm_policy_queue_process(struct timer_list *t);
 159
 160 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
 161 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
 162                                                 int dir);
 163
 164 static struct xfrm_pol_inexact_bin *
 165 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
 166                            u32 if_id);
 167
 168 static struct xfrm_pol_inexact_bin *
 169 xfrm_policy_inexact_lookup_rcu(struct net *net,
 170                                u8 type, u16 family, u8 dir, u32 if_id);
 171 static struct xfrm_policy *
 172 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
 173                         bool excl);
 174 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
 175                                             struct xfrm_policy *policy);
 176
 177 static bool
 178 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
 179                                     struct xfrm_pol_inexact_bin *b,
 180                                     const xfrm_address_t *saddr,
 181                                     const xfrm_address_t *daddr);
 182
 183 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
 184 {
 185         return refcount_inc_not_zero(&policy->refcnt);
 186 }
 187
 188 static inline bool
 189 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 190 {
 191         const struct flowi4 *fl4 = &fl->u.ip4;
 192
 193         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
 194                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
 195                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
 196                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
 197                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
 198                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
 199 }
 200
 201 static inline bool
 202 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 203 {
 204         const struct flowi6 *fl6 = &fl->u.ip6;
 205
 206         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
 207                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
 208                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
 209                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
 210                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
 211                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
 212 }
 213
 214 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
 215                          unsigned short family)
 216 {
 217         switch (family) {
 218         case AF_INET:
 219                 return __xfrm4_selector_match(sel, fl);
 220         case AF_INET6:
 221                 return __xfrm6_selector_match(sel, fl);
 222         }
 223         return false;
 224 }
 225
 226 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 227 {
 228         const struct xfrm_policy_afinfo *afinfo;
 229
 230         if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
 231                 return NULL;
 232         rcu_read_lock();
 233         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 234         if (unlikely(!afinfo))
 235                 rcu_read_unlock();
 236         return afinfo;
 237 }
 238
 239 /* Called with rcu_read_lock(). */
 240 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
 241 {
 242         return rcu_dereference(xfrm_if_cb);
 243 }
 244
 245 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
 246                                     const xfrm_address_t *saddr,
 247                                     const xfrm_address_t *daddr,
 248                                     int family, u32 mark)
 249 {
 250         const struct xfrm_policy_afinfo *afinfo;
 251         struct dst_entry *dst;
 252
 253         afinfo = xfrm_policy_get_afinfo(family);
 254         if (unlikely(afinfo == NULL))
 255                 return ERR_PTR(-EAFNOSUPPORT);
 256
 257         dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
 258
 259         rcu_read_unlock();
 260
 261         return dst;
 262 }
 263 EXPORT_SYMBOL(__xfrm_dst_lookup);
 264
 265 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
 266                                                 int tos, int oif,
 267                                                 xfrm_address_t *prev_saddr,
 268                                                 xfrm_address_t *prev_daddr,
 269                                                 int family, u32 mark)
 270 {
 271         struct net *net = xs_net(x);
 272         xfrm_address_t *saddr = &x->props.saddr;
 273         xfrm_address_t *daddr = &x->id.daddr;
 274         struct dst_entry *dst;
 275
 276         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 277                 saddr = x->coaddr;
 278                 daddr = prev_daddr;
 279         }
 280         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 281                 saddr = prev_saddr;
 282                 daddr = x->coaddr;
 283         }
 284
 285         dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
 286
 287         if (!IS_ERR(dst)) {
 288                 if (prev_saddr != saddr)
 289                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 290                 if (prev_daddr != daddr)
 291                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 292         }
 293
 294         return dst;
 295 }
 296
 297 static inline unsigned long make_jiffies(long secs)
 298 {
 299         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 300                 return MAX_SCHEDULE_TIMEOUT-1;
 301         else
 302                 return secs*HZ;
 303 }
 304
 305 static void xfrm_policy_timer(struct timer_list *t)
 306 {
 307         struct xfrm_policy *xp = from_timer(xp, t, timer);
 308         time64_t now = ktime_get_real_seconds();
 309         time64_t next = TIME64_MAX;
 310         int warn = 0;
 311         int dir;
 312
 313         read_lock(&xp->lock);
 314
 315         if (unlikely(xp->walk.dead))
 316                 goto out;
 317
 318         dir = xfrm_policy_id2dir(xp->index);
 319
 320         if (xp->lft.hard_add_expires_seconds) {
 321                 time64_t tmo = xp->lft.hard_add_expires_seconds +
 322                         xp->curlft.add_time - now;
 323                 if (tmo <= 0)
 324                         goto expired;
 325                 if (tmo < next)
 326                         next = tmo;
 327         }
 328         if (xp->lft.hard_use_expires_seconds) {
 329                 time64_t tmo = xp->lft.hard_use_expires_seconds +
 330                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 331                 if (tmo <= 0)
 332                         goto expired;
 333                 if (tmo < next)
 334                         next = tmo;
 335         }
 336         if (xp->lft.soft_add_expires_seconds) {
 337                 time64_t tmo = xp->lft.soft_add_expires_seconds +
 338                         xp->curlft.add_time - now;
 339                 if (tmo <= 0) {
 340                         warn = 1;
 341                         tmo = XFRM_KM_TIMEOUT;
 342                 }
 343                 if (tmo < next)
 344                         next = tmo;
 345         }
 346         if (xp->lft.soft_use_expires_seconds) {
 347                 time64_t tmo = xp->lft.soft_use_expires_seconds +
 348                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 349                 if (tmo <= 0) {
 350                         warn = 1;
 351                         tmo = XFRM_KM_TIMEOUT;
 352                 }
 353                 if (tmo < next)
 354                         next = tmo;
 355         }
 356
 357         if (warn)
 358                 km_policy_expired(xp, dir, 0, 0);
 359         if (next != TIME64_MAX &&
 360             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 361                 xfrm_pol_hold(xp);
 362
 363 out:
 364         read_unlock(&xp->lock);
 365         xfrm_pol_put(xp);
 366         return;
 367
 368 expired:
 369         read_unlock(&xp->lock);
 370         if (!xfrm_policy_delete(xp, dir))
 371                 km_policy_expired(xp, dir, 1, 0);
 372         xfrm_pol_put(xp);
 373 }
 374
 375 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 376  * SPD calls.
 377  */
 378
 379 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 380 {
 381         struct xfrm_policy *policy;
 382
 383         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 384
 385         if (policy) {
 386                 write_pnet(&policy->xp_net, net);
 387                 INIT_LIST_HEAD(&policy->walk.all);
 388                 INIT_HLIST_NODE(&policy->bydst_inexact_list);
 389                 INIT_HLIST_NODE(&policy->bydst);
 390                 INIT_HLIST_NODE(&policy->byidx);
 391                 rwlock_init(&policy->lock);
 392                 refcount_set(&policy->refcnt, 1);
 393                 skb_queue_head_init(&policy->polq.hold_queue);
 394                 timer_setup(&policy->timer, xfrm_policy_timer, 0);
 395                 timer_setup(&policy->polq.hold_timer,
 396                             xfrm_policy_queue_process, 0);
 397         }
 398         return policy;
 399 }
 400 EXPORT_SYMBOL(xfrm_policy_alloc);
 401
 402 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
 403 {
 404         struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
 405
 406         security_xfrm_policy_free(policy->security);
 407         kfree(policy);
 408 }
 409
 410 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 411
 412 void xfrm_policy_destroy(struct xfrm_policy *policy)
 413 {
 414         BUG_ON(!policy->walk.dead);
 415
 416         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 417                 BUG();
 418
 419         call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
 420 }
 421 EXPORT_SYMBOL(xfrm_policy_destroy);
 422
 423 /* Rule must be locked. Release descendant resources, announce
 424  * entry dead. The rule must be unlinked from lists to the moment.
 425  */
 426
 427 static void xfrm_policy_kill(struct xfrm_policy *policy)
 428 {
 429         policy->walk.dead = 1;
 430
 431         atomic_inc(&policy->genid);
 432
 433         if (del_timer(&policy->polq.hold_timer))
 434                 xfrm_pol_put(policy);
 435         skb_queue_purge(&policy->polq.hold_queue);
 436
 437         if (del_timer(&policy->timer))
 438                 xfrm_pol_put(policy);
 439
 440         xfrm_pol_put(policy);
 441 }
 442
 443 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 444
 445 static inline unsigned int idx_hash(struct net *net, u32 index)
 446 {
 447         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 448 }
 449
 450 /* calculate policy hash thresholds */
 451 static void __get_hash_thresh(struct net *net,
 452                               unsigned short family, int dir,
 453                               u8 *dbits, u8 *sbits)
 454 {
 455         switch (family) {
 456         case AF_INET:
 457                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 458                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 459                 break;
 460
 461         case AF_INET6:
 462                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 463                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 464                 break;
 465
 466         default:
 467                 *dbits = 0;
 468                 *sbits = 0;
 469         }
 470 }
 471
 472 static struct hlist_head *policy_hash_bysel(struct net *net,
 473                                             const struct xfrm_selector *sel,
 474                                             unsigned short family, int dir)
 475 {
 476         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 477         unsigned int hash;
 478         u8 dbits;
 479         u8 sbits;
 480
 481         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 482         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 483
 484         if (hash == hmask + 1)
 485                 return NULL;
 486
 487         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 488                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 489 }
 490
 491 static struct hlist_head *policy_hash_direct(struct net *net,
 492                                              const xfrm_address_t *daddr,
 493                                              const xfrm_address_t *saddr,
 494                                              unsigned short family, int dir)
 495 {
 496         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 497         unsigned int hash;
 498         u8 dbits;
 499         u8 sbits;
 500
 501         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 502         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 503
 504         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 505                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 506 }
 507
 508 static void xfrm_dst_hash_transfer(struct net *net,
 509                                    struct hlist_head *list,
 510                                    struct hlist_head *ndsttable,
 511                                    unsigned int nhashmask,
 512                                    int dir)
 513 {
 514         struct hlist_node *tmp, *entry0 = NULL;
 515         struct xfrm_policy *pol;
 516         unsigned int h0 = 0;
 517         u8 dbits;
 518         u8 sbits;
 519
 520 redo:
 521         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 522                 unsigned int h;
 523
 524                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 525                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 526                                 pol->family, nhashmask, dbits, sbits);
 527                 if (!entry0) {
 528                         hlist_del_rcu(&pol->bydst);
 529                         hlist_add_head_rcu(&pol->bydst, ndsttable + h);
 530                         h0 = h;
 531                 } else {
 532                         if (h != h0)
 533                                 continue;
 534                         hlist_del_rcu(&pol->bydst);
 535                         hlist_add_behind_rcu(&pol->bydst, entry0);
 536                 }
 537                 entry0 = &pol->bydst;
 538         }
 539         if (!hlist_empty(list)) {
 540                 entry0 = NULL;
 541                 goto redo;
 542         }
 543 }
 544
 545 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 546                                    struct hlist_head *nidxtable,
 547                                    unsigned int nhashmask)
 548 {
 549         struct hlist_node *tmp;
 550         struct xfrm_policy *pol;
 551
 552         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 553                 unsigned int h;
 554
 555                 h = __idx_hash(pol->index, nhashmask);
 556                 hlist_add_head(&pol->byidx, nidxtable+h);
 557         }
 558 }
 559
 560 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 561 {
 562         return ((old_hmask + 1) << 1) - 1;
 563 }
 564
 565 static void xfrm_bydst_resize(struct net *net, int dir)
 566 {
 567         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 568         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 569         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 570         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 571         struct hlist_head *odst;
 572         int i;
 573
 574         if (!ndst)
 575                 return;
 576
 577         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 578         write_seqcount_begin(&xfrm_policy_hash_generation);
 579
 580         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 581                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 582
 583         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 584                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 585
 586         for (i = hmask; i >= 0; i--)
 587                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 588
 589         rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
 590         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 591
 592         write_seqcount_end(&xfrm_policy_hash_generation);
 593         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 594
 595         synchronize_rcu();
 596
 597         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 598 }
 599
 600 static void xfrm_byidx_resize(struct net *net, int total)
 601 {
 602         unsigned int hmask = net->xfrm.policy_idx_hmask;
 603         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 604         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 605         struct hlist_head *oidx = net->xfrm.policy_byidx;
 606         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 607         int i;
 608
 609         if (!nidx)
 610                 return;
 611
 612         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 613
 614         for (i = hmask; i >= 0; i--)
 615                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 616
 617         net->xfrm.policy_byidx = nidx;
 618         net->xfrm.policy_idx_hmask = nhashmask;
 619
 620         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 621
 622         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 623 }
 624
 625 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 626 {
 627         unsigned int cnt = net->xfrm.policy_count[dir];
 628         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 629
 630         if (total)
 631                 *total += cnt;
 632
 633         if ((hmask + 1) < xfrm_policy_hashmax &&
 634             cnt > hmask)
 635                 return 1;
 636
 637         return 0;
 638 }
 639
 640 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 641 {
 642         unsigned int hmask = net->xfrm.policy_idx_hmask;
 643
 644         if ((hmask + 1) < xfrm_policy_hashmax &&
 645             total > hmask)
 646                 return 1;
 647
 648         return 0;
 649 }
 650
 651 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 652 {
 653         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 654         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 655         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 656         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 657         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 658         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 659         si->spdhcnt = net->xfrm.policy_idx_hmask;
 660         si->spdhmcnt = xfrm_policy_hashmax;
 661 }
 662 EXPORT_SYMBOL(xfrm_spd_getinfo);
 663
 664 static DEFINE_MUTEX(hash_resize_mutex);
 665 static void xfrm_hash_resize(struct work_struct *work)
 666 {
 667         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 668         int dir, total;
 669
 670         mutex_lock(&hash_resize_mutex);
 671
 672         total = 0;
 673         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 674                 if (xfrm_bydst_should_resize(net, dir, &total))
 675                         xfrm_bydst_resize(net, dir);
 676         }
 677         if (xfrm_byidx_should_resize(net, total))
 678                 xfrm_byidx_resize(net, total);
 679
 680         mutex_unlock(&hash_resize_mutex);
 681 }
 682
 683 static void xfrm_hash_reset_inexact_table(struct net *net)
 684 {
 685         struct xfrm_pol_inexact_bin *b;
 686
 687         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
 688
 689         list_for_each_entry(b, &net->xfrm.inexact_bins, inexact_bins)
 690                 INIT_HLIST_HEAD(&b->hhead);
 691 }
 692
 693 /* Make sure *pol can be inserted into fastbin.
 694  * Useful to check that later insert requests will be sucessful
 695  * (provided xfrm_policy_lock is held throughout).
 696  */
 697 static struct xfrm_pol_inexact_bin *
 698 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
 699 {
 700         struct xfrm_pol_inexact_bin *bin, *prev;
 701         struct xfrm_pol_inexact_key k = {
 702                 .family = pol->family,
 703                 .type = pol->type,
 704                 .dir = dir,
 705                 .if_id = pol->if_id,
 706         };
 707         struct net *net = xp_net(pol);
 708
 709         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
 710
 711         write_pnet(&k.net, net);
 712         bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
 713                                      xfrm_pol_inexact_params);
 714         if (bin)
 715                 return bin;
 716
 717         bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
 718         if (!bin)
 719                 return NULL;
 720
 721         bin->k = k;
 722         INIT_HLIST_HEAD(&bin->hhead);
 723         bin->root_d = RB_ROOT;
 724         bin->root_s = RB_ROOT;
 725         seqcount_init(&bin->count);
 726
 727         prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
 728                                                 &bin->k, &bin->head,
 729                                                 xfrm_pol_inexact_params);
 730         if (!prev) {
 731                 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
 732                 return bin;
 733         }
 734
 735         kfree(bin);
 736
 737         return IS_ERR(prev) ? NULL : prev;
 738 }
 739
 740 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
 741                                                int family, u8 prefixlen)
 742 {
 743         if (xfrm_addr_any(addr, family))
 744                 return true;
 745
 746         if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
 747                 return true;
 748
 749         if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
 750                 return true;
 751
 752         return false;
 753 }
 754
 755 static bool
 756 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
 757 {
 758         const xfrm_address_t *addr;
 759         bool saddr_any, daddr_any;
 760         u8 prefixlen;
 761
 762         addr = &policy->selector.saddr;
 763         prefixlen = policy->selector.prefixlen_s;
 764
 765         saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 766                                                        policy->family,
 767                                                        prefixlen);
 768         addr = &policy->selector.daddr;
 769         prefixlen = policy->selector.prefixlen_d;
 770         daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 771                                                        policy->family,
 772                                                        prefixlen);
 773         return saddr_any && daddr_any;
 774 }
 775
 776 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
 777                                        const xfrm_address_t *addr, u8 prefixlen)
 778 {
 779         node->addr = *addr;
 780         node->prefixlen = prefixlen;
 781 }
 782
 783 static struct xfrm_pol_inexact_node *
 784 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
 785 {
 786         struct xfrm_pol_inexact_node *node;
 787
 788         node = kzalloc(sizeof(*node), GFP_ATOMIC);
 789         if (node)
 790                 xfrm_pol_inexact_node_init(node, addr, prefixlen);
 791
 792         return node;
 793 }
 794
 795 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
 796                                   const xfrm_address_t *b,
 797                                   u8 prefixlen, u16 family)
 798 {
 799         unsigned int pdw, pbi;
 800         int delta = 0;
 801
 802         switch (family) {
 803         case AF_INET:
 804                 if (sizeof(long) == 4 && prefixlen == 0)
 805                         return ntohl(a->a4) - ntohl(b->a4);
 806                 return (ntohl(a->a4) & ((~0UL << (32 - prefixlen)))) -
 807                        (ntohl(b->a4) & ((~0UL << (32 - prefixlen))));
 808         case AF_INET6:
 809                 pdw = prefixlen >> 5;
 810                 pbi = prefixlen & 0x1f;
 811
 812                 if (pdw) {
 813                         delta = memcmp(a->a6, b->a6, pdw << 2);
 814                         if (delta)
 815                                 return delta;
 816                 }
 817                 if (pbi) {
 818                         u32 mask = ~0u << (32 - pbi);
 819
 820                         delta = (ntohl(a->a6[pdw]) & mask) -
 821                                 (ntohl(b->a6[pdw]) & mask);
 822                 }
 823                 break;
 824         default:
 825                 break;
 826         }
 827
 828         return delta;
 829 }
 830
 831 static void xfrm_policy_inexact_list_reinsert(struct net *net,
 832                                               struct xfrm_pol_inexact_node *n,
 833                                               u16 family)
 834 {
 835         unsigned int matched_s, matched_d;
 836         struct hlist_node *newpos = NULL;
 837         struct xfrm_policy *policy, *p;
 838
 839         matched_s = 0;
 840         matched_d = 0;
 841
 842         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 843                 bool matches_s, matches_d;
 844
 845                 if (!policy->bydst_reinsert)
 846                         continue;
 847
 848                 WARN_ON_ONCE(policy->family != family);
 849
 850                 policy->bydst_reinsert = false;
 851                 hlist_for_each_entry(p, &n->hhead, bydst) {
 852                         if (policy->priority >= p->priority)
 853                                 newpos = &p->bydst;
 854                         else
 855                                 break;
 856                 }
 857
 858                 if (newpos)
 859                         hlist_add_behind(&policy->bydst, newpos);
 860                 else
 861                         hlist_add_head(&policy->bydst, &n->hhead);
 862
 863                 /* paranoia checks follow.
 864                  * Check that the reinserted policy matches at least
 865                  * saddr or daddr for current node prefix.
 866                  *
 867                  * Matching both is fine, matching saddr in one policy
 868                  * (but not daddr) and then matching only daddr in another
 869                  * is a bug.
 870                  */
 871                 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
 872                                                    &n->addr,
 873                                                    n->prefixlen,
 874                                                    family) == 0;
 875                 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
 876                                                    &n->addr,
 877                                                    n->prefixlen,
 878                                                    family) == 0;
 879                 if (matches_s && matches_d)
 880                         continue;
 881
 882                 WARN_ON_ONCE(!matches_s && !matches_d);
 883                 if (matches_s)
 884                         matched_s++;
 885                 if (matches_d)
 886                         matched_d++;
 887                 WARN_ON_ONCE(matched_s && matched_d);
 888         }
 889 }
 890
 891 static void xfrm_policy_inexact_node_reinsert(struct net *net,
 892                                               struct xfrm_pol_inexact_node *n,
 893                                               struct rb_root *new,
 894                                               u16 family)
 895 {
 896         struct rb_node **p, *parent = NULL;
 897         struct xfrm_pol_inexact_node *node;
 898
 899         /* we should not have another subtree here */
 900         WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
 901
 902         p = &new->rb_node;
 903         while (*p) {
 904                 u8 prefixlen;
 905                 int delta;
 906
 907                 parent = *p;
 908                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
 909
 910                 prefixlen = min(node->prefixlen, n->prefixlen);
 911
 912                 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
 913                                                prefixlen, family);
 914                 if (delta < 0) {
 915                         p = &parent->rb_left;
 916                 } else if (delta > 0) {
 917                         p = &parent->rb_right;
 918                 } else {
 919                         struct xfrm_policy *tmp;
 920
 921                         hlist_for_each_entry(tmp, &node->hhead, bydst)
 922                                 tmp->bydst_reinsert = true;
 923                         hlist_for_each_entry(tmp, &n->hhead, bydst)
 924                                 tmp->bydst_reinsert = true;
 925
 926                         INIT_HLIST_HEAD(&node->hhead);
 927                         xfrm_policy_inexact_list_reinsert(net, node, family);
 928
 929                         if (node->prefixlen == n->prefixlen) {
 930                                 kfree_rcu(n, rcu);
 931                                 return;
 932                         }
 933
 934                         rb_erase(*p, new);
 935                         kfree_rcu(n, rcu);
 936                         n = node;
 937                         n->prefixlen = prefixlen;
 938                         *p = new->rb_node;
 939                         parent = NULL;
 940                 }
 941         }
 942
 943         rb_link_node_rcu(&n->node, parent, p);
 944         rb_insert_color(&n->node, new);
 945 }
 946
 947 /* merge nodes v and n */
 948 static void xfrm_policy_inexact_node_merge(struct net *net,
 949                                            struct xfrm_pol_inexact_node *v,
 950                                            struct xfrm_pol_inexact_node *n,
 951                                            u16 family)
 952 {
 953         struct xfrm_pol_inexact_node *node;
 954         struct xfrm_policy *tmp;
 955         struct rb_node *rnode;
 956
 957         /* To-be-merged node v has a subtree.
 958          *
 959          * Dismantle it and insert its nodes to n->root.
 960          */
 961         while ((rnode = rb_first(&v->root)) != NULL) {
 962                 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
 963                 rb_erase(&node->node, &v->root);
 964                 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
 965                                                   family);
 966         }
 967
 968         hlist_for_each_entry(tmp, &v->hhead, bydst)
 969                 tmp->bydst_reinsert = true;
 970         hlist_for_each_entry(tmp, &n->hhead, bydst)
 971                 tmp->bydst_reinsert = true;
 972
 973         INIT_HLIST_HEAD(&n->hhead);
 974         xfrm_policy_inexact_list_reinsert(net, n, family);
 975 }
 976
 977 static struct xfrm_pol_inexact_node *
 978 xfrm_policy_inexact_insert_node(struct net *net,
 979                                 struct rb_root *root,
 980                                 xfrm_address_t *addr,
 981                                 u16 family, u8 prefixlen, u8 dir)
 982 {
 983         struct xfrm_pol_inexact_node *cached = NULL;
 984         struct rb_node **p, *parent = NULL;
 985         struct xfrm_pol_inexact_node *node;
 986
 987         p = &root->rb_node;
 988         while (*p) {
 989                 int delta;
 990
 991                 parent = *p;
 992                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
 993
 994                 delta = xfrm_policy_addr_delta(addr, &node->addr,
 995                                                node->prefixlen,
 996                                                family);
 997                 if (delta == 0 && prefixlen >= node->prefixlen) {
 998                         WARN_ON_ONCE(cached); /* ipsec policies got lost */
 999                         return node;
1000                 }
1001
1002                 if (delta < 0)
1003                         p = &parent->rb_left;
1004                 else
1005                         p = &parent->rb_right;
1006
1007                 if (prefixlen < node->prefixlen) {
1008                         delta = xfrm_policy_addr_delta(addr, &node->addr,
1009                                                        prefixlen,
1010                                                        family);
1011                         if (delta)
1012                                 continue;
1013
1014                         /* This node is a subnet of the new prefix. It needs
1015                          * to be removed and re-inserted with the smaller
1016                          * prefix and all nodes that are now also covered
1017                          * by the reduced prefixlen.
1018                          */
1019                         rb_erase(&node->node, root);
1020
1021                         if (!cached) {
1022                                 xfrm_pol_inexact_node_init(node, addr,
1023                                                            prefixlen);
1024                                 cached = node;
1025                         } else {
1026                                 /* This node also falls within the new
1027                                  * prefixlen. Merge the to-be-reinserted
1028                                  * node and this one.
1029                                  */
1030                                 xfrm_policy_inexact_node_merge(net, node,
1031                                                                cached, family);
1032                                 kfree_rcu(node, rcu);
1033                         }
1034
1035                         /* restart */
1036                         p = &root->rb_node;
1037                         parent = NULL;
1038                 }
1039         }
1040
1041         node = cached;
1042         if (!node) {
1043                 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1044                 if (!node)
1045                         return NULL;
1046         }
1047
1048         rb_link_node_rcu(&node->node, parent, p);
1049         rb_insert_color(&node->node, root);
1050
1051         return node;
1052 }
1053
1054 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1055 {
1056         struct xfrm_pol_inexact_node *node;
1057         struct rb_node *rn = rb_first(r);
1058
1059         while (rn) {
1060                 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1061
1062                 xfrm_policy_inexact_gc_tree(&node->root, rm);
1063                 rn = rb_next(rn);
1064
1065                 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1066                         WARN_ON_ONCE(rm);
1067                         continue;
1068                 }
1069
1070                 rb_erase(&node->node, r);
1071                 kfree_rcu(node, rcu);
1072         }
1073 }
1074
1075 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1076 {
1077         write_seqcount_begin(&b->count);
1078         xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1079         xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1080         write_seqcount_end(&b->count);
1081
1082         if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1083             !hlist_empty(&b->hhead)) {
1084                 WARN_ON_ONCE(net_exit);
1085                 return;
1086         }
1087
1088         if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1089                                    xfrm_pol_inexact_params) == 0) {
1090                 list_del(&b->inexact_bins);
1091                 kfree_rcu(b, rcu);
1092         }
1093 }
1094
1095 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1096 {
1097         struct net *net = read_pnet(&b->k.net);
1098
1099         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1100         __xfrm_policy_inexact_prune_bin(b, false);
1101         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1102 }
1103
1104 static void __xfrm_policy_inexact_flush(struct net *net)
1105 {
1106         struct xfrm_pol_inexact_bin *bin, *t;
1107
1108         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1109
1110         list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1111                 __xfrm_policy_inexact_prune_bin(bin, false);
1112 }
1113
1114 static struct hlist_head *
1115 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1116                                 struct xfrm_policy *policy, u8 dir)
1117 {
1118         struct xfrm_pol_inexact_node *n;
1119         struct net *net;
1120
1121         net = xp_net(policy);
1122         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1123
1124         if (xfrm_policy_inexact_insert_use_any_list(policy))
1125                 return &bin->hhead;
1126
1127         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1128                                                policy->family,
1129                                                policy->selector.prefixlen_d)) {
1130                 write_seqcount_begin(&bin->count);
1131                 n = xfrm_policy_inexact_insert_node(net,
1132                                                     &bin->root_s,
1133                                                     &policy->selector.saddr,
1134                                                     policy->family,
1135                                                     policy->selector.prefixlen_s,
1136                                                     dir);
1137                 write_seqcount_end(&bin->count);
1138                 if (!n)
1139                         return NULL;
1140
1141                 return &n->hhead;
1142         }
1143
1144         /* daddr is fixed */
1145         write_seqcount_begin(&bin->count);
1146         n = xfrm_policy_inexact_insert_node(net,
1147                                             &bin->root_d,
1148                                             &policy->selector.daddr,
1149                                             policy->family,
1150                                             policy->selector.prefixlen_d, dir);
1151         write_seqcount_end(&bin->count);
1152         if (!n)
1153                 return NULL;
1154
1155         /* saddr is wildcard */
1156         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1157                                                policy->family,
1158                                                policy->selector.prefixlen_s))
1159                 return &n->hhead;
1160
1161         write_seqcount_begin(&bin->count);
1162         n = xfrm_policy_inexact_insert_node(net,
1163                                             &n->root,
1164                                             &policy->selector.saddr,
1165                                             policy->family,
1166                                             policy->selector.prefixlen_s, dir);
1167         write_seqcount_end(&bin->count);
1168         if (!n)
1169                 return NULL;
1170
1171         return &n->hhead;
1172 }
1173
1174 static struct xfrm_policy *
1175 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1176 {
1177         struct xfrm_pol_inexact_bin *bin;
1178         struct xfrm_policy *delpol;
1179         struct hlist_head *chain;
1180         struct net *net;
1181
1182         bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1183         if (!bin)
1184                 return ERR_PTR(-ENOMEM);
1185
1186         net = xp_net(policy);
1187         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1188
1189         chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1190         if (!chain) {
1191                 __xfrm_policy_inexact_prune_bin(bin, false);
1192                 return ERR_PTR(-ENOMEM);
1193         }
1194
1195         delpol = xfrm_policy_insert_list(chain, policy, excl);
1196         if (delpol && excl) {
1197                 __xfrm_policy_inexact_prune_bin(bin, false);
1198                 return ERR_PTR(-EEXIST);
1199         }
1200
1201         chain = &net->xfrm.policy_inexact[dir];
1202         xfrm_policy_insert_inexact_list(chain, policy);
1203
1204         if (delpol)
1205                 __xfrm_policy_inexact_prune_bin(bin, false);
1206
1207         return delpol;
1208 }
1209
1210 static void xfrm_hash_rebuild(struct work_struct *work)
1211 {
1212         struct net *net = container_of(work, struct net,
1213                                        xfrm.policy_hthresh.work);
1214         unsigned int hmask;
1215         struct xfrm_policy *pol;
1216         struct xfrm_policy *policy;
1217         struct hlist_head *chain;
1218         struct hlist_head *odst;
1219         struct hlist_node *newpos;
1220         int i;
1221         int dir;
1222         unsigned seq;
1223         u8 lbits4, rbits4, lbits6, rbits6;
1224
1225         mutex_lock(&hash_resize_mutex);
1226
1227         /* read selector prefixlen thresholds */
1228         do {
1229                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1230
1231                 lbits4 = net->xfrm.policy_hthresh.lbits4;
1232                 rbits4 = net->xfrm.policy_hthresh.rbits4;
1233                 lbits6 = net->xfrm.policy_hthresh.lbits6;
1234                 rbits6 = net->xfrm.policy_hthresh.rbits6;
1235         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1236
1237         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1238
1239         /* make sure that we can insert the indirect policies again before
1240          * we start with destructive action.
1241          */
1242         list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1243                 struct xfrm_pol_inexact_bin *bin;
1244                 u8 dbits, sbits;
1245
1246                 dir = xfrm_policy_id2dir(policy->index);
1247                 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1248                         continue;
1249
1250                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1251                         if (policy->family == AF_INET) {
1252                                 dbits = rbits4;
1253                                 sbits = lbits4;
1254                         } else {
1255                                 dbits = rbits6;
1256                                 sbits = lbits6;
1257                         }
1258                 } else {
1259                         if (policy->family == AF_INET) {
1260                                 dbits = lbits4;
1261                                 sbits = rbits4;
1262                         } else {
1263                                 dbits = lbits6;
1264                                 sbits = rbits6;
1265                         }
1266                 }
1267
1268                 if (policy->selector.prefixlen_d < dbits ||
1269                     policy->selector.prefixlen_s < sbits)
1270                         continue;
1271
1272                 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1273                 if (!bin)
1274                         goto out_unlock;
1275
1276                 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1277                         goto out_unlock;
1278         }
1279
1280         /* reset the bydst and inexact table in all directions */
1281         xfrm_hash_reset_inexact_table(net);
1282
1283         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1284                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
1285                 hmask = net->xfrm.policy_bydst[dir].hmask;
1286                 odst = net->xfrm.policy_bydst[dir].table;
1287                 for (i = hmask; i >= 0; i--)
1288                         INIT_HLIST_HEAD(odst + i);
1289                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1290                         /* dir out => dst = remote, src = local */
1291                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1292                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1293                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1294                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1295                 } else {
1296                         /* dir in/fwd => dst = local, src = remote */
1297                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1298                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1299                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1300                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1301                 }
1302         }
1303
1304         /* re-insert all policies by order of creation */
1305         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1306                 if (policy->walk.dead)
1307                         continue;
1308                 dir = xfrm_policy_id2dir(policy->index);
1309                 if (dir >= XFRM_POLICY_MAX) {
1310                         /* skip socket policies */
1311                         continue;
1312                 }
1313                 newpos = NULL;
1314                 chain = policy_hash_bysel(net, &policy->selector,
1315                                           policy->family, dir);
1316                 if (!chain) {
1317                         void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1318
1319                         WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1320                         continue;
1321                 }
1322
1323                 hlist_for_each_entry(pol, chain, bydst) {
1324                         if (policy->priority >= pol->priority)
1325                                 newpos = &pol->bydst;
1326                         else
1327                                 break;
1328                 }
1329                 if (newpos)
1330                         hlist_add_behind_rcu(&policy->bydst, newpos);
1331                 else
1332                         hlist_add_head_rcu(&policy->bydst, chain);
1333         }
1334
1335 out_unlock:
1336         __xfrm_policy_inexact_flush(net);
1337         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1338
1339         mutex_unlock(&hash_resize_mutex);
1340 }
1341
1342 void xfrm_policy_hash_rebuild(struct net *net)
1343 {
1344         schedule_work(&net->xfrm.policy_hthresh.work);
1345 }
1346 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1347
1348 /* Generate new index... KAME seems to generate them ordered by cost
1349  * of an absolute inpredictability of ordering of rules. This will not pass. */
1350 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1351 {
1352         static u32 idx_generator;
1353
1354         for (;;) {
1355                 struct hlist_head *list;
1356                 struct xfrm_policy *p;
1357                 u32 idx;
1358                 int found;
1359
1360                 if (!index) {
1361                         idx = (idx_generator | dir);
1362                         idx_generator += 8;
1363                 } else {
1364                         idx = index;
1365                         index = 0;
1366                 }
1367
1368                 if (idx == 0)
1369                         idx = 8;
1370                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1371                 found = 0;
1372                 hlist_for_each_entry(p, list, byidx) {
1373                         if (p->index == idx) {
1374                                 found = 1;
1375                                 break;
1376                         }
1377                 }
1378                 if (!found)
1379                         return idx;
1380         }
1381 }
1382
1383 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1384 {
1385         u32 *p1 = (u32 *) s1;
1386         u32 *p2 = (u32 *) s2;
1387         int len = sizeof(struct xfrm_selector) / sizeof(u32);
1388         int i;
1389
1390         for (i = 0; i < len; i++) {
1391                 if (p1[i] != p2[i])
1392                         return 1;
1393         }
1394
1395         return 0;
1396 }
1397
1398 static void xfrm_policy_requeue(struct xfrm_policy *old,
1399                                 struct xfrm_policy *new)
1400 {
1401         struct xfrm_policy_queue *pq = &old->polq;
1402         struct sk_buff_head list;
1403
1404         if (skb_queue_empty(&pq->hold_queue))
1405                 return;
1406
1407         __skb_queue_head_init(&list);
1408
1409         spin_lock_bh(&pq->hold_queue.lock);
1410         skb_queue_splice_init(&pq->hold_queue, &list);
1411         if (del_timer(&pq->hold_timer))
1412                 xfrm_pol_put(old);
1413         spin_unlock_bh(&pq->hold_queue.lock);
1414
1415         pq = &new->polq;
1416
1417         spin_lock_bh(&pq->hold_queue.lock);
1418         skb_queue_splice(&list, &pq->hold_queue);
1419         pq->timeout = XFRM_QUEUE_TMO_MIN;
1420         if (!mod_timer(&pq->hold_timer, jiffies))
1421                 xfrm_pol_hold(new);
1422         spin_unlock_bh(&pq->hold_queue.lock);
1423 }
1424
1425 static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
1426                                    struct xfrm_policy *pol)
1427 {
1428         u32 mark = policy->mark.v & policy->mark.m;
1429
1430         if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
1431                 return true;
1432
1433         if ((mark & pol->mark.m) == pol->mark.v &&
1434             policy->priority == pol->priority)
1435                 return true;
1436
1437         return false;
1438 }
1439
1440 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1441 {
1442         const struct xfrm_pol_inexact_key *k = data;
1443         u32 a = k->type << 24 | k->dir << 16 | k->family;
1444
1445         return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1446                             seed);
1447 }
1448
1449 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1450 {
1451         const struct xfrm_pol_inexact_bin *b = data;
1452
1453         return xfrm_pol_bin_key(&b->k, 0, seed);
1454 }
1455
1456 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1457                             const void *ptr)
1458 {
1459         const struct xfrm_pol_inexact_key *key = arg->key;
1460         const struct xfrm_pol_inexact_bin *b = ptr;
1461         int ret;
1462
1463         if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1464                 return -1;
1465
1466         ret = b->k.dir ^ key->dir;
1467         if (ret)
1468                 return ret;
1469
1470         ret = b->k.type ^ key->type;
1471         if (ret)
1472                 return ret;
1473
1474         ret = b->k.family ^ key->family;
1475         if (ret)
1476                 return ret;
1477
1478         return b->k.if_id ^ key->if_id;
1479 }
1480
1481 static const struct rhashtable_params xfrm_pol_inexact_params = {
1482         .head_offset            = offsetof(struct xfrm_pol_inexact_bin, head),
1483         .hashfn                 = xfrm_pol_bin_key,
1484         .obj_hashfn             = xfrm_pol_bin_obj,
1485         .obj_cmpfn              = xfrm_pol_bin_cmp,
1486         .automatic_shrinking    = true,
1487 };
1488
1489 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1490                                             struct xfrm_policy *policy)
1491 {
1492         struct xfrm_policy *pol, *delpol = NULL;
1493         struct hlist_node *newpos = NULL;
1494         int i = 0;
1495
1496         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1497                 if (pol->type == policy->type &&
1498                     pol->if_id == policy->if_id &&
1499                     !selector_cmp(&pol->selector, &policy->selector) &&
1500                     xfrm_policy_mark_match(policy, pol) &&
1501                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1502                     !WARN_ON(delpol)) {
1503                         delpol = pol;
1504                         if (policy->priority > pol->priority)
1505                                 continue;
1506                 } else if (policy->priority >= pol->priority) {
1507                         newpos = &pol->bydst_inexact_list;
1508                         continue;
1509                 }
1510                 if (delpol)
1511                         break;
1512         }
1513
1514         if (newpos)
1515                 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1516         else
1517                 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1518
1519         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1520                 pol->pos = i;
1521                 i++;
1522         }
1523 }
1524
1525 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1526                                                    struct xfrm_policy *policy,
1527                                                    bool excl)
1528 {
1529         struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1530
1531         hlist_for_each_entry(pol, chain, bydst) {
1532                 if (pol->type == policy->type &&
1533                     pol->if_id == policy->if_id &&
1534                     !selector_cmp(&pol->selector, &policy->selector) &&
1535                     xfrm_policy_mark_match(policy, pol) &&
1536                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1537                     !WARN_ON(delpol)) {
1538                         if (excl)
1539                                 return ERR_PTR(-EEXIST);
1540                         delpol = pol;
1541                         if (policy->priority > pol->priority)
1542                                 continue;
1543                 } else if (policy->priority >= pol->priority) {
1544                         newpos = pol;
1545                         continue;
1546                 }
1547                 if (delpol)
1548                         break;
1549         }
1550
1551         if (newpos)
1552                 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1553         else
1554                 hlist_add_head_rcu(&policy->bydst, chain);
1555
1556         return delpol;
1557 }
1558
1559 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1560 {
1561         struct net *net = xp_net(policy);
1562         struct xfrm_policy *delpol;
1563         struct hlist_head *chain;
1564
1565         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1566         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1567         if (chain)
1568                 delpol = xfrm_policy_insert_list(chain, policy, excl);
1569         else
1570                 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1571
1572         if (IS_ERR(delpol)) {
1573                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1574                 return PTR_ERR(delpol);
1575         }
1576
1577         __xfrm_policy_link(policy, dir);
1578
1579         /* After previous checking, family can either be AF_INET or AF_INET6 */
1580         if (policy->family == AF_INET)
1581                 rt_genid_bump_ipv4(net);
1582         else
1583                 rt_genid_bump_ipv6(net);
1584
1585         if (delpol) {
1586                 xfrm_policy_requeue(delpol, policy);
1587                 __xfrm_policy_unlink(delpol, dir);
1588         }
1589         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1590         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1591         policy->curlft.add_time = ktime_get_real_seconds();
1592         policy->curlft.use_time = 0;
1593         if (!mod_timer(&policy->timer, jiffies + HZ))
1594                 xfrm_pol_hold(policy);
1595         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1596
1597         if (delpol)
1598                 xfrm_policy_kill(delpol);
1599         else if (xfrm_bydst_should_resize(net, dir, NULL))
1600                 schedule_work(&net->xfrm.policy_hash_work);
1601
1602         return 0;
1603 }
1604 EXPORT_SYMBOL(xfrm_policy_insert);
1605
1606 static struct xfrm_policy *
1607 __xfrm_policy_bysel_ctx(struct hlist_head *chain, u32 mark, u32 if_id,
1608                         u8 type, int dir,
1609                         struct xfrm_selector *sel,
1610                         struct xfrm_sec_ctx *ctx)
1611 {
1612         struct xfrm_policy *pol;
1613
1614         if (!chain)
1615                 return NULL;
1616
1617         hlist_for_each_entry(pol, chain, bydst) {
1618                 if (pol->type == type &&
1619                     pol->if_id == if_id &&
1620                     (mark & pol->mark.m) == pol->mark.v &&
1621                     !selector_cmp(sel, &pol->selector) &&
1622                     xfrm_sec_ctx_match(ctx, pol->security))
1623                         return pol;
1624         }
1625
1626         return NULL;
1627 }
1628
1629 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u32 if_id,
1630                                           u8 type, int dir,
1631                                           struct xfrm_selector *sel,
1632                                           struct xfrm_sec_ctx *ctx, int delete,
1633                                           int *err)
1634 {
1635         struct xfrm_pol_inexact_bin *bin = NULL;
1636         struct xfrm_policy *pol, *ret = NULL;
1637         struct hlist_head *chain;
1638
1639         *err = 0;
1640         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1641         chain = policy_hash_bysel(net, sel, sel->family, dir);
1642         if (!chain) {
1643                 struct xfrm_pol_inexact_candidates cand;
1644                 int i;
1645
1646                 bin = xfrm_policy_inexact_lookup(net, type,
1647                                                  sel->family, dir, if_id);
1648                 if (!bin) {
1649                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1650                         return NULL;
1651                 }
1652
1653                 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1654                                                          &sel->saddr,
1655                                                          &sel->daddr)) {
1656                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1657                         return NULL;
1658                 }
1659
1660                 pol = NULL;
1661                 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1662                         struct xfrm_policy *tmp;
1663
1664                         tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1665                                                       if_id, type, dir,
1666                                                       sel, ctx);
1667                         if (!tmp)
1668                                 continue;
1669
1670                         if (!pol || tmp->pos < pol->pos)
1671                                 pol = tmp;
1672                 }
1673         } else {
1674                 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1675                                               sel, ctx);
1676         }
1677
1678         if (pol) {
1679                 xfrm_pol_hold(pol);
1680                 if (delete) {
1681                         *err = security_xfrm_policy_delete(pol->security);
1682                         if (*err) {
1683                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1684                                 return pol;
1685                         }
1686                         __xfrm_policy_unlink(pol, dir);
1687                 }
1688                 ret = pol;
1689         }
1690         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1691
1692         if (ret && delete)
1693                 xfrm_policy_kill(ret);
1694         if (bin && delete)
1695                 xfrm_policy_inexact_prune_bin(bin);
1696         return ret;
1697 }
1698 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1699
1700 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u32 if_id,
1701                                      u8 type, int dir, u32 id, int delete,
1702                                      int *err)
1703 {
1704         struct xfrm_policy *pol, *ret;
1705         struct hlist_head *chain;
1706
1707         *err = -ENOENT;
1708         if (xfrm_policy_id2dir(id) != dir)
1709                 return NULL;
1710
1711         *err = 0;
1712         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1713         chain = net->xfrm.policy_byidx + idx_hash(net, id);
1714         ret = NULL;
1715         hlist_for_each_entry(pol, chain, byidx) {
1716                 if (pol->type == type && pol->index == id &&
1717                     pol->if_id == if_id &&
1718                     (mark & pol->mark.m) == pol->mark.v) {
1719                         xfrm_pol_hold(pol);
1720                         if (delete) {
1721                                 *err = security_xfrm_policy_delete(
1722                                                                 pol->security);
1723                                 if (*err) {
1724                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1725                                         return pol;
1726                                 }
1727                                 __xfrm_policy_unlink(pol, dir);
1728                         }
1729                         ret = pol;
1730                         break;
1731                 }
1732         }
1733         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1734
1735         if (ret && delete)
1736                 xfrm_policy_kill(ret);
1737         return ret;
1738 }
1739 EXPORT_SYMBOL(xfrm_policy_byid);
1740
1741 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1742 static inline int
1743 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1744 {
1745         struct xfrm_policy *pol;
1746         int err = 0;
1747
1748         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1749                 if (pol->walk.dead ||
1750                     xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1751                     pol->type != type)
1752                         continue;
1753
1754                 err = security_xfrm_policy_delete(pol->security);
1755                 if (err) {
1756                         xfrm_audit_policy_delete(pol, 0, task_valid);
1757                         return err;
1758                 }
1759         }
1760         return err;
1761 }
1762 #else
1763 static inline int
1764 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1765 {
1766         return 0;
1767 }
1768 #endif
1769
1770 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1771 {
1772         int dir, err = 0, cnt = 0;
1773         struct xfrm_policy *pol;
1774
1775         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1776
1777         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1778         if (err)
1779                 goto out;
1780
1781 again:
1782         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1783                 dir = xfrm_policy_id2dir(pol->index);
1784                 if (pol->walk.dead ||
1785                     dir >= XFRM_POLICY_MAX ||
1786                     pol->type != type)
1787                         continue;
1788
1789                 __xfrm_policy_unlink(pol, dir);
1790                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1791                 cnt++;
1792                 xfrm_audit_policy_delete(pol, 1, task_valid);
1793                 xfrm_policy_kill(pol);
1794                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1795                 goto again;
1796         }
1797         if (cnt)
1798                 __xfrm_policy_inexact_flush(net);
1799         else
1800                 err = -ESRCH;
1801 out:
1802         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1803         return err;
1804 }
1805 EXPORT_SYMBOL(xfrm_policy_flush);
1806
1807 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1808                      int (*func)(struct xfrm_policy *, int, int, void*),
1809                      void *data)
1810 {
1811         struct xfrm_policy *pol;
1812         struct xfrm_policy_walk_entry *x;
1813         int error = 0;
1814
1815         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1816             walk->type != XFRM_POLICY_TYPE_ANY)
1817                 return -EINVAL;
1818
1819         if (list_empty(&walk->walk.all) && walk->seq != 0)
1820                 return 0;
1821
1822         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1823         if (list_empty(&walk->walk.all))
1824                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1825         else
1826                 x = list_first_entry(&walk->walk.all,
1827                                      struct xfrm_policy_walk_entry, all);
1828
1829         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1830                 if (x->dead)
1831                         continue;
1832                 pol = container_of(x, struct xfrm_policy, walk);
1833                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1834                     walk->type != pol->type)
1835                         continue;
1836                 error = func(pol, xfrm_policy_id2dir(pol->index),
1837                              walk->seq, data);
1838                 if (error) {
1839                         list_move_tail(&walk->walk.all, &x->all);
1840                         goto out;
1841                 }
1842                 walk->seq++;
1843         }
1844         if (walk->seq == 0) {
1845                 error = -ENOENT;
1846                 goto out;
1847         }
1848         list_del_init(&walk->walk.all);
1849 out:
1850         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1851         return error;
1852 }
1853 EXPORT_SYMBOL(xfrm_policy_walk);
1854
1855 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1856 {
1857         INIT_LIST_HEAD(&walk->walk.all);
1858         walk->walk.dead = 1;
1859         walk->type = type;
1860         walk->seq = 0;
1861 }
1862 EXPORT_SYMBOL(xfrm_policy_walk_init);
1863
1864 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1865 {
1866         if (list_empty(&walk->walk.all))
1867                 return;
1868
1869         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1870         list_del(&walk->walk.all);
1871         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1872 }
1873 EXPORT_SYMBOL(xfrm_policy_walk_done);
1874
1875 /*
1876  * Find policy to apply to this flow.
1877  *
1878  * Returns 0 if policy found, else an -errno.
1879  */
1880 static int xfrm_policy_match(const struct xfrm_policy *pol,
1881                              const struct flowi *fl,
1882                              u8 type, u16 family, int dir, u32 if_id)
1883 {
1884         const struct xfrm_selector *sel = &pol->selector;
1885         int ret = -ESRCH;
1886         bool match;
1887
1888         if (pol->family != family ||
1889             pol->if_id != if_id ||
1890             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1891             pol->type != type)
1892                 return ret;
1893
1894         match = xfrm_selector_match(sel, fl, family);
1895         if (match)
1896                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1897                                                   dir);
1898         return ret;
1899 }
1900
1901 static struct xfrm_pol_inexact_node *
1902 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1903                                 seqcount_t *count,
1904                                 const xfrm_address_t *addr, u16 family)
1905 {
1906         const struct rb_node *parent;
1907         int seq;
1908
1909 again:
1910         seq = read_seqcount_begin(count);
1911
1912         parent = rcu_dereference_raw(r->rb_node);
1913         while (parent) {
1914                 struct xfrm_pol_inexact_node *node;
1915                 int delta;
1916
1917                 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1918
1919                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1920                                                node->prefixlen, family);
1921                 if (delta < 0) {
1922                         parent = rcu_dereference_raw(parent->rb_left);
1923                         continue;
1924                 } else if (delta > 0) {
1925                         parent = rcu_dereference_raw(parent->rb_right);
1926                         continue;
1927                 }
1928
1929                 return node;
1930         }
1931
1932         if (read_seqcount_retry(count, seq))
1933                 goto again;
1934
1935         return NULL;
1936 }
1937
1938 static bool
1939 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1940                                     struct xfrm_pol_inexact_bin *b,
1941                                     const xfrm_address_t *saddr,
1942                                     const xfrm_address_t *daddr)
1943 {
1944         struct xfrm_pol_inexact_node *n;
1945         u16 family;
1946
1947         if (!b)
1948                 return false;
1949
1950         family = b->k.family;
1951         memset(cand, 0, sizeof(*cand));
1952         cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1953
1954         n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1955                                             family);
1956         if (n) {
1957                 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1958                 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1959                                                     family);
1960                 if (n)
1961                         cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1962         }
1963
1964         n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1965                                             family);
1966         if (n)
1967                 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1968
1969         return true;
1970 }
1971
1972 static struct xfrm_pol_inexact_bin *
1973 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1974                                u8 dir, u32 if_id)
1975 {
1976         struct xfrm_pol_inexact_key k = {
1977                 .family = family,
1978                 .type = type,
1979                 .dir = dir,
1980                 .if_id = if_id,
1981         };
1982
1983         write_pnet(&k.net, net);
1984
1985         return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1986                                  xfrm_pol_inexact_params);
1987 }
1988
1989 static struct xfrm_pol_inexact_bin *
1990 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1991                            u8 dir, u32 if_id)
1992 {
1993         struct xfrm_pol_inexact_bin *bin;
1994
1995         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1996
1997         rcu_read_lock();
1998         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
1999         rcu_read_unlock();
2000
2001         return bin;
2002 }
2003
2004 static struct xfrm_policy *
2005 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2006                               struct xfrm_policy *prefer,
2007                               const struct flowi *fl,
2008                               u8 type, u16 family, int dir, u32 if_id)
2009 {
2010         u32 priority = prefer ? prefer->priority : ~0u;
2011         struct xfrm_policy *pol;
2012
2013         if (!chain)
2014                 return NULL;
2015
2016         hlist_for_each_entry_rcu(pol, chain, bydst) {
2017                 int err;
2018
2019                 if (pol->priority > priority)
2020                         break;
2021
2022                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2023                 if (err) {
2024                         if (err != -ESRCH)
2025                                 return ERR_PTR(err);
2026
2027                         continue;
2028                 }
2029
2030                 if (prefer) {
2031                         /* matches.  Is it older than *prefer? */
2032                         if (pol->priority == priority &&
2033                             prefer->pos < pol->pos)
2034                                 return prefer;
2035                 }
2036
2037                 return pol;
2038         }
2039
2040         return NULL;
2041 }
2042
2043 static struct xfrm_policy *
2044 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2045                             struct xfrm_policy *prefer,
2046                             const struct flowi *fl,
2047                             u8 type, u16 family, int dir, u32 if_id)
2048 {
2049         struct xfrm_policy *tmp;
2050         int i;
2051
2052         for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2053                 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2054                                                     prefer,
2055                                                     fl, type, family, dir,
2056                                                     if_id);
2057                 if (!tmp)
2058                         continue;
2059
2060                 if (IS_ERR(tmp))
2061                         return tmp;
2062                 prefer = tmp;
2063         }
2064
2065         return prefer;
2066 }
2067
2068 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2069                                                      const struct flowi *fl,
2070                                                      u16 family, u8 dir,
2071                                                      u32 if_id)
2072 {
2073         struct xfrm_pol_inexact_candidates cand;
2074         const xfrm_address_t *daddr, *saddr;
2075         struct xfrm_pol_inexact_bin *bin;
2076         struct xfrm_policy *pol, *ret;
2077         struct hlist_head *chain;
2078         unsigned int sequence;
2079         u32 priority;
2080         int err;
2081
2082         daddr = xfrm_flowi_daddr(fl, family);
2083         saddr = xfrm_flowi_saddr(fl, family);
2084         if (unlikely(!daddr || !saddr))
2085                 return NULL;
2086
2087         rcu_read_lock();
2088  retry:
2089         do {
2090                 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2091                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2092         } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
2093
2094         priority = ~0U;
2095         ret = NULL;
2096         hlist_for_each_entry_rcu(pol, chain, bydst) {
2097                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2098                 if (err) {
2099                         if (err == -ESRCH)
2100                                 continue;
2101                         else {
2102                                 ret = ERR_PTR(err);
2103                                 goto fail;
2104                         }
2105                 } else {
2106                         ret = pol;
2107                         priority = ret->priority;
2108                         break;
2109                 }
2110         }
2111         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2112         if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2113                                                          daddr))
2114                 goto skip_inexact;
2115
2116         pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2117                                           family, dir, if_id);
2118         if (pol) {
2119                 ret = pol;
2120                 if (IS_ERR(pol))
2121                         goto fail;
2122         }
2123
2124 skip_inexact:
2125         if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
2126                 goto retry;
2127
2128         if (ret && !xfrm_pol_hold_rcu(ret))
2129                 goto retry;
2130 fail:
2131         rcu_read_unlock();
2132
2133         return ret;
2134 }
2135
2136 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2137                                               const struct flowi *fl,
2138                                               u16 family, u8 dir, u32 if_id)
2139 {
2140 #ifdef CONFIG_XFRM_SUB_POLICY
2141         struct xfrm_policy *pol;
2142
2143         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2144                                         dir, if_id);
2145         if (pol != NULL)
2146                 return pol;
2147 #endif
2148         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2149                                          dir, if_id);
2150 }
2151
2152 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2153                                                  const struct flowi *fl,
2154                                                  u16 family, u32 if_id)
2155 {
2156         struct xfrm_policy *pol;
2157
2158         rcu_read_lock();
2159  again:
2160         pol = rcu_dereference(sk->sk_policy[dir]);
2161         if (pol != NULL) {
2162                 bool match;
2163                 int err = 0;
2164
2165                 if (pol->family != family) {
2166                         pol = NULL;
2167                         goto out;
2168                 }
2169
2170                 match = xfrm_selector_match(&pol->selector, fl, family);
2171                 if (match) {
2172                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2173                             pol->if_id != if_id) {
2174                                 pol = NULL;
2175                                 goto out;
2176                         }
2177                         err = security_xfrm_policy_lookup(pol->security,
2178                                                       fl->flowi_secid,
2179                                                       dir);
2180                         if (!err) {
2181                                 if (!xfrm_pol_hold_rcu(pol))
2182                                         goto again;
2183                         } else if (err == -ESRCH) {
2184                                 pol = NULL;
2185                         } else {
2186                                 pol = ERR_PTR(err);
2187                         }
2188                 } else
2189                         pol = NULL;
2190         }
2191 out:
2192         rcu_read_unlock();
2193         return pol;
2194 }
2195
2196 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2197 {
2198         struct net *net = xp_net(pol);
2199
2200         list_add(&pol->walk.all, &net->xfrm.policy_all);
2201         net->xfrm.policy_count[dir]++;
2202         xfrm_pol_hold(pol);
2203 }
2204
2205 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2206                                                 int dir)
2207 {
2208         struct net *net = xp_net(pol);
2209
2210         if (list_empty(&pol->walk.all))
2211                 return NULL;
2212
2213         /* Socket policies are not hashed. */
2214         if (!hlist_unhashed(&pol->bydst)) {
2215                 hlist_del_rcu(&pol->bydst);
2216                 hlist_del_init(&pol->bydst_inexact_list);
2217                 hlist_del(&pol->byidx);
2218         }
2219
2220         list_del_init(&pol->walk.all);
2221         net->xfrm.policy_count[dir]--;
2222
2223         return pol;
2224 }
2225
2226 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2227 {
2228         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2229 }
2230
2231 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2232 {
2233         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2234 }
2235
2236 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2237 {
2238         struct net *net = xp_net(pol);
2239
2240         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2241         pol = __xfrm_policy_unlink(pol, dir);
2242         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2243         if (pol) {
2244                 xfrm_policy_kill(pol);
2245                 return 0;
2246         }
2247         return -ENOENT;
2248 }
2249 EXPORT_SYMBOL(xfrm_policy_delete);
2250
2251 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2252 {
2253         struct net *net = sock_net(sk);
2254         struct xfrm_policy *old_pol;
2255
2256 #ifdef CONFIG_XFRM_SUB_POLICY
2257         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2258                 return -EINVAL;
2259 #endif
2260
2261         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2262         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2263                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2264         if (pol) {
2265                 pol->curlft.add_time = ktime_get_real_seconds();
2266                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2267                 xfrm_sk_policy_link(pol, dir);
2268         }
2269         rcu_assign_pointer(sk->sk_policy[dir], pol);
2270         if (old_pol) {
2271                 if (pol)
2272                         xfrm_policy_requeue(old_pol, pol);
2273
2274                 /* Unlinking succeeds always. This is the only function
2275                  * allowed to delete or replace socket policy.
2276                  */
2277                 xfrm_sk_policy_unlink(old_pol, dir);
2278         }
2279         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2280
2281         if (old_pol) {
2282                 xfrm_policy_kill(old_pol);
2283         }
2284         return 0;
2285 }
2286
2287 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2288 {
2289         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2290         struct net *net = xp_net(old);
2291
2292         if (newp) {
2293                 newp->selector = old->selector;
2294                 if (security_xfrm_policy_clone(old->security,
2295                                                &newp->security)) {
2296                         kfree(newp);
2297                         return NULL;  /* ENOMEM */
2298                 }
2299                 newp->lft = old->lft;
2300                 newp->curlft = old->curlft;
2301                 newp->mark = old->mark;
2302                 newp->if_id = old->if_id;
2303                 newp->action = old->action;
2304                 newp->flags = old->flags;
2305                 newp->xfrm_nr = old->xfrm_nr;
2306                 newp->index = old->index;
2307                 newp->type = old->type;
2308                 newp->family = old->family;
2309                 memcpy(newp->xfrm_vec, old->xfrm_vec,
2310                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2311                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2312                 xfrm_sk_policy_link(newp, dir);
2313                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2314                 xfrm_pol_put(newp);
2315         }
2316         return newp;
2317 }
2318
2319 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2320 {
2321         const struct xfrm_policy *p;
2322         struct xfrm_policy *np;
2323         int i, ret = 0;
2324
2325         rcu_read_lock();
2326         for (i = 0; i < 2; i++) {
2327                 p = rcu_dereference(osk->sk_policy[i]);
2328                 if (p) {
2329                         np = clone_policy(p, i);
2330                         if (unlikely(!np)) {
2331                                 ret = -ENOMEM;
2332                                 break;
2333                         }
2334                         rcu_assign_pointer(sk->sk_policy[i], np);
2335                 }
2336         }
2337         rcu_read_unlock();
2338         return ret;
2339 }
2340
2341 static int
2342 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2343                xfrm_address_t *remote, unsigned short family, u32 mark)
2344 {
2345         int err;
2346         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2347
2348         if (unlikely(afinfo == NULL))
2349                 return -EINVAL;
2350         err = afinfo->get_saddr(net, oif, local, remote, mark);
2351         rcu_read_unlock();
2352         return err;
2353 }
2354
2355 /* Resolve list of templates for the flow, given policy. */
2356
2357 static int
2358 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2359                       struct xfrm_state **xfrm, unsigned short family)
2360 {
2361         struct net *net = xp_net(policy);
2362         int nx;
2363         int i, error;
2364         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2365         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2366         xfrm_address_t tmp;
2367
2368         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2369                 struct xfrm_state *x;
2370                 xfrm_address_t *remote = daddr;
2371                 xfrm_address_t *local  = saddr;
2372                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2373
2374                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2375                     tmpl->mode == XFRM_MODE_BEET) {
2376                         remote = &tmpl->id.daddr;
2377                         local = &tmpl->saddr;
2378                         if (xfrm_addr_any(local, tmpl->encap_family)) {
2379                                 error = xfrm_get_saddr(net, fl->flowi_oif,
2380                                                        &tmp, remote,
2381                                                        tmpl->encap_family, 0);
2382                                 if (error)
2383                                         goto fail;
2384                                 local = &tmp;
2385                         }
2386                 }
2387
2388                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2389                                     family, policy->if_id);
2390
2391                 if (x && x->km.state == XFRM_STATE_VALID) {
2392                         xfrm[nx++] = x;
2393                         daddr = remote;
2394                         saddr = local;
2395                         continue;
2396                 }
2397                 if (x) {
2398                         error = (x->km.state == XFRM_STATE_ERROR ?
2399                                  -EINVAL : -EAGAIN);
2400                         xfrm_state_put(x);
2401                 } else if (error == -ESRCH) {
2402                         error = -EAGAIN;
2403                 }
2404
2405                 if (!tmpl->optional)
2406                         goto fail;
2407         }
2408         return nx;
2409
2410 fail:
2411         for (nx--; nx >= 0; nx--)
2412                 xfrm_state_put(xfrm[nx]);
2413         return error;
2414 }
2415
2416 static int
2417 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2418                   struct xfrm_state **xfrm, unsigned short family)
2419 {
2420         struct xfrm_state *tp[XFRM_MAX_DEPTH];
2421         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2422         int cnx = 0;
2423         int error;
2424         int ret;
2425         int i;
2426
2427         for (i = 0; i < npols; i++) {
2428                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2429                         error = -ENOBUFS;
2430                         goto fail;
2431                 }
2432
2433                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2434                 if (ret < 0) {
2435                         error = ret;
2436                         goto fail;
2437                 } else
2438                         cnx += ret;
2439         }
2440
2441         /* found states are sorted for outbound processing */
2442         if (npols > 1)
2443                 xfrm_state_sort(xfrm, tpp, cnx, family);
2444
2445         return cnx;
2446
2447  fail:
2448         for (cnx--; cnx >= 0; cnx--)
2449                 xfrm_state_put(tpp[cnx]);
2450         return error;
2451
2452 }
2453
2454 static int xfrm_get_tos(const struct flowi *fl, int family)
2455 {
2456         const struct xfrm_policy_afinfo *afinfo;
2457         int tos;
2458
2459         afinfo = xfrm_policy_get_afinfo(family);
2460         if (!afinfo)
2461                 return 0;
2462
2463         tos = afinfo->get_tos(fl);
2464
2465         rcu_read_unlock();
2466
2467         return tos;
2468 }
2469
2470 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2471 {
2472         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2473         struct dst_ops *dst_ops;
2474         struct xfrm_dst *xdst;
2475
2476         if (!afinfo)
2477                 return ERR_PTR(-EINVAL);
2478
2479         switch (family) {
2480         case AF_INET:
2481                 dst_ops = &net->xfrm.xfrm4_dst_ops;
2482                 break;
2483 #if IS_ENABLED(CONFIG_IPV6)
2484         case AF_INET6:
2485                 dst_ops = &net->xfrm.xfrm6_dst_ops;
2486                 break;
2487 #endif
2488         default:
2489                 BUG();
2490         }
2491         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2492
2493         if (likely(xdst)) {
2494                 struct dst_entry *dst = &xdst->u.dst;
2495
2496                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2497         } else
2498                 xdst = ERR_PTR(-ENOBUFS);
2499
2500         rcu_read_unlock();
2501
2502         return xdst;
2503 }
2504
2505 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2506                                  int nfheader_len)
2507 {
2508         const struct xfrm_policy_afinfo *afinfo =
2509                 xfrm_policy_get_afinfo(dst->ops->family);
2510         int err;
2511
2512         if (!afinfo)
2513                 return -EINVAL;
2514
2515         err = afinfo->init_path(path, dst, nfheader_len);
2516
2517         rcu_read_unlock();
2518
2519         return err;
2520 }
2521
2522 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2523                                 const struct flowi *fl)
2524 {
2525         const struct xfrm_policy_afinfo *afinfo =
2526                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2527         int err;
2528
2529         if (!afinfo)
2530                 return -EINVAL;
2531
2532         err = afinfo->fill_dst(xdst, dev, fl);
2533
2534         rcu_read_unlock();
2535
2536         return err;
2537 }
2538
2539
2540 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2541  * all the metrics... Shortly, bundle a bundle.
2542  */
2543
2544 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2545                                             struct xfrm_state **xfrm,
2546                                             struct xfrm_dst **bundle,
2547                                             int nx,
2548                                             const struct flowi *fl,
2549                                             struct dst_entry *dst)
2550 {
2551         struct net *net = xp_net(policy);
2552         unsigned long now = jiffies;
2553         struct net_device *dev;
2554         struct xfrm_mode *inner_mode;
2555         struct xfrm_dst *xdst_prev = NULL;
2556         struct xfrm_dst *xdst0 = NULL;
2557         int i = 0;
2558         int err;
2559         int header_len = 0;
2560         int nfheader_len = 0;
2561         int trailer_len = 0;
2562         int tos;
2563         int family = policy->selector.family;
2564         xfrm_address_t saddr, daddr;
2565
2566         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2567
2568         tos = xfrm_get_tos(fl, family);
2569
2570         dst_hold(dst);
2571
2572         for (; i < nx; i++) {
2573                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2574                 struct dst_entry *dst1 = &xdst->u.dst;
2575
2576                 err = PTR_ERR(xdst);
2577                 if (IS_ERR(xdst)) {
2578                         dst_release(dst);
2579                         goto put_states;
2580                 }
2581
2582                 bundle[i] = xdst;
2583                 if (!xdst_prev)
2584                         xdst0 = xdst;
2585                 else
2586                         /* Ref count is taken during xfrm_alloc_dst()
2587                          * No need to do dst_clone() on dst1
2588                          */
2589                         xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2590
2591                 if (xfrm[i]->sel.family == AF_UNSPEC) {
2592                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
2593                                                         xfrm_af2proto(family));
2594                         if (!inner_mode) {
2595                                 err = -EAFNOSUPPORT;
2596                                 dst_release(dst);
2597                                 goto put_states;
2598                         }
2599                 } else
2600                         inner_mode = xfrm[i]->inner_mode;
2601
2602                 xdst->route = dst;
2603                 dst_copy_metrics(dst1, dst);
2604
2605                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2606                         __u32 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2607
2608                         family = xfrm[i]->props.family;
2609                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2610                                               &saddr, &daddr, family, mark);
2611                         err = PTR_ERR(dst);
2612                         if (IS_ERR(dst))
2613                                 goto put_states;
2614                 } else
2615                         dst_hold(dst);
2616
2617                 dst1->xfrm = xfrm[i];
2618                 xdst->xfrm_genid = xfrm[i]->genid;
2619
2620                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2621                 dst1->flags |= DST_HOST;
2622                 dst1->lastuse = now;
2623
2624                 dst1->input = dst_discard;
2625                 dst1->output = inner_mode->afinfo->output;
2626
2627                 xdst_prev = xdst;
2628
2629                 header_len += xfrm[i]->props.header_len;
2630                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2631                         nfheader_len += xfrm[i]->props.header_len;
2632                 trailer_len += xfrm[i]->props.trailer_len;
2633         }
2634
2635         xfrm_dst_set_child(xdst_prev, dst);
2636         xdst0->path = dst;
2637
2638         err = -ENODEV;
2639         dev = dst->dev;
2640         if (!dev)
2641                 goto free_dst;
2642
2643         xfrm_init_path(xdst0, dst, nfheader_len);
2644         xfrm_init_pmtu(bundle, nx);
2645
2646         for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2647              xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2648                 err = xfrm_fill_dst(xdst_prev, dev, fl);
2649                 if (err)
2650                         goto free_dst;
2651
2652                 xdst_prev->u.dst.header_len = header_len;
2653                 xdst_prev->u.dst.trailer_len = trailer_len;
2654                 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2655                 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2656         }
2657
2658         return &xdst0->u.dst;
2659
2660 put_states:
2661         for (; i < nx; i++)
2662                 xfrm_state_put(xfrm[i]);
2663 free_dst:
2664         if (xdst0)
2665                 dst_release_immediate(&xdst0->u.dst);
2666
2667         return ERR_PTR(err);
2668 }
2669
2670 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2671                                 struct xfrm_policy **pols,
2672                                 int *num_pols, int *num_xfrms)
2673 {
2674         int i;
2675
2676         if (*num_pols == 0 || !pols[0]) {
2677                 *num_pols = 0;
2678                 *num_xfrms = 0;
2679                 return 0;
2680         }
2681         if (IS_ERR(pols[0]))
2682                 return PTR_ERR(pols[0]);
2683
2684         *num_xfrms = pols[0]->xfrm_nr;
2685
2686 #ifdef CONFIG_XFRM_SUB_POLICY
2687         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2688             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2689                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2690                                                     XFRM_POLICY_TYPE_MAIN,
2691                                                     fl, family,
2692                                                     XFRM_POLICY_OUT,
2693                                                     pols[0]->if_id);
2694                 if (pols[1]) {
2695                         if (IS_ERR(pols[1])) {
2696                                 xfrm_pols_put(pols, *num_pols);
2697                                 return PTR_ERR(pols[1]);
2698                         }
2699                         (*num_pols)++;
2700                         (*num_xfrms) += pols[1]->xfrm_nr;
2701                 }
2702         }
2703 #endif
2704         for (i = 0; i < *num_pols; i++) {
2705                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2706                         *num_xfrms = -1;
2707                         break;
2708                 }
2709         }
2710
2711         return 0;
2712
2713 }
2714
2715 static struct xfrm_dst *
2716 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2717                                const struct flowi *fl, u16 family,
2718                                struct dst_entry *dst_orig)
2719 {
2720         struct net *net = xp_net(pols[0]);
2721         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2722         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2723         struct xfrm_dst *xdst;
2724         struct dst_entry *dst;
2725         int err;
2726
2727         /* Try to instantiate a bundle */
2728         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2729         if (err <= 0) {
2730                 if (err == 0)
2731                         return NULL;
2732
2733                 if (err != -EAGAIN)
2734                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2735                 return ERR_PTR(err);
2736         }
2737
2738         dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2739         if (IS_ERR(dst)) {
2740                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2741                 return ERR_CAST(dst);
2742         }
2743
2744         xdst = (struct xfrm_dst *)dst;
2745         xdst->num_xfrms = err;
2746         xdst->num_pols = num_pols;
2747         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2748         xdst->policy_genid = atomic_read(&pols[0]->genid);
2749
2750         return xdst;
2751 }
2752
2753 static void xfrm_policy_queue_process(struct timer_list *t)
2754 {
2755         struct sk_buff *skb;
2756         struct sock *sk;
2757         struct dst_entry *dst;
2758         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2759         struct net *net = xp_net(pol);
2760         struct xfrm_policy_queue *pq = &pol->polq;
2761         struct flowi fl;
2762         struct sk_buff_head list;
2763
2764         spin_lock(&pq->hold_queue.lock);
2765         skb = skb_peek(&pq->hold_queue);
2766         if (!skb) {
2767                 spin_unlock(&pq->hold_queue.lock);
2768                 goto out;
2769         }
2770         dst = skb_dst(skb);
2771         sk = skb->sk;
2772         xfrm_decode_session(skb, &fl, dst->ops->family);
2773         spin_unlock(&pq->hold_queue.lock);
2774
2775         dst_hold(xfrm_dst_path(dst));
2776         dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2777         if (IS_ERR(dst))
2778                 goto purge_queue;
2779
2780         if (dst->flags & DST_XFRM_QUEUE) {
2781                 dst_release(dst);
2782
2783                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2784                         goto purge_queue;
2785
2786                 pq->timeout = pq->timeout << 1;
2787                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2788                         xfrm_pol_hold(pol);
2789                 goto out;
2790         }
2791
2792         dst_release(dst);
2793
2794         __skb_queue_head_init(&list);
2795
2796         spin_lock(&pq->hold_queue.lock);
2797         pq->timeout = 0;
2798         skb_queue_splice_init(&pq->hold_queue, &list);
2799         spin_unlock(&pq->hold_queue.lock);
2800
2801         while (!skb_queue_empty(&list)) {
2802                 skb = __skb_dequeue(&list);
2803
2804                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2805                 dst_hold(xfrm_dst_path(skb_dst(skb)));
2806                 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2807                 if (IS_ERR(dst)) {
2808                         kfree_skb(skb);
2809                         continue;
2810                 }
2811
2812                 nf_reset(skb);
2813                 skb_dst_drop(skb);
2814                 skb_dst_set(skb, dst);
2815
2816                 dst_output(net, skb->sk, skb);
2817         }
2818
2819 out:
2820         xfrm_pol_put(pol);
2821         return;
2822
2823 purge_queue:
2824         pq->timeout = 0;
2825         skb_queue_purge(&pq->hold_queue);
2826         xfrm_pol_put(pol);
2827 }
2828
2829 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2830 {
2831         unsigned long sched_next;
2832         struct dst_entry *dst = skb_dst(skb);
2833         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2834         struct xfrm_policy *pol = xdst->pols[0];
2835         struct xfrm_policy_queue *pq = &pol->polq;
2836
2837         if (unlikely(skb_fclone_busy(sk, skb))) {
2838                 kfree_skb(skb);
2839                 return 0;
2840         }
2841
2842         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2843                 kfree_skb(skb);
2844                 return -EAGAIN;
2845         }
2846
2847         skb_dst_force(skb);
2848
2849         spin_lock_bh(&pq->hold_queue.lock);
2850
2851         if (!pq->timeout)
2852                 pq->timeout = XFRM_QUEUE_TMO_MIN;
2853
2854         sched_next = jiffies + pq->timeout;
2855
2856         if (del_timer(&pq->hold_timer)) {
2857                 if (time_before(pq->hold_timer.expires, sched_next))
2858                         sched_next = pq->hold_timer.expires;
2859                 xfrm_pol_put(pol);
2860         }
2861
2862         __skb_queue_tail(&pq->hold_queue, skb);
2863         if (!mod_timer(&pq->hold_timer, sched_next))
2864                 xfrm_pol_hold(pol);
2865
2866         spin_unlock_bh(&pq->hold_queue.lock);
2867
2868         return 0;
2869 }
2870
2871 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2872                                                  struct xfrm_flo *xflo,
2873                                                  const struct flowi *fl,
2874                                                  int num_xfrms,
2875                                                  u16 family)
2876 {
2877         int err;
2878         struct net_device *dev;
2879         struct dst_entry *dst;
2880         struct dst_entry *dst1;
2881         struct xfrm_dst *xdst;
2882
2883         xdst = xfrm_alloc_dst(net, family);
2884         if (IS_ERR(xdst))
2885                 return xdst;
2886
2887         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2888             net->xfrm.sysctl_larval_drop ||
2889             num_xfrms <= 0)
2890                 return xdst;
2891
2892         dst = xflo->dst_orig;
2893         dst1 = &xdst->u.dst;
2894         dst_hold(dst);
2895         xdst->route = dst;
2896
2897         dst_copy_metrics(dst1, dst);
2898
2899         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2900         dst1->flags |= DST_HOST | DST_XFRM_QUEUE;
2901         dst1->lastuse = jiffies;
2902
2903         dst1->input = dst_discard;
2904         dst1->output = xdst_queue_output;
2905
2906         dst_hold(dst);
2907         xfrm_dst_set_child(xdst, dst);
2908         xdst->path = dst;
2909
2910         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2911
2912         err = -ENODEV;
2913         dev = dst->dev;
2914         if (!dev)
2915                 goto free_dst;
2916
2917         err = xfrm_fill_dst(xdst, dev, fl);
2918         if (err)
2919                 goto free_dst;
2920
2921 out:
2922         return xdst;
2923
2924 free_dst:
2925         dst_release(dst1);
2926         xdst = ERR_PTR(err);
2927         goto out;
2928 }
2929
2930 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2931                                            const struct flowi *fl,
2932                                            u16 family, u8 dir,
2933                                            struct xfrm_flo *xflo, u32 if_id)
2934 {
2935         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2936         int num_pols = 0, num_xfrms = 0, err;
2937         struct xfrm_dst *xdst;
2938
2939         /* Resolve policies to use if we couldn't get them from
2940          * previous cache entry */
2941         num_pols = 1;
2942         pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2943         err = xfrm_expand_policies(fl, family, pols,
2944                                            &num_pols, &num_xfrms);
2945         if (err < 0)
2946                 goto inc_error;
2947         if (num_pols == 0)
2948                 return NULL;
2949         if (num_xfrms <= 0)
2950                 goto make_dummy_bundle;
2951
2952         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2953                                               xflo->dst_orig);
2954         if (IS_ERR(xdst)) {
2955                 err = PTR_ERR(xdst);
2956                 if (err == -EREMOTE) {
2957                         xfrm_pols_put(pols, num_pols);
2958                         return NULL;
2959                 }
2960
2961                 if (err != -EAGAIN)
2962                         goto error;
2963                 goto make_dummy_bundle;
2964         } else if (xdst == NULL) {
2965                 num_xfrms = 0;
2966                 goto make_dummy_bundle;
2967         }
2968
2969         return xdst;
2970
2971 make_dummy_bundle:
2972         /* We found policies, but there's no bundles to instantiate:
2973          * either because the policy blocks, has no transformations or
2974          * we could not build template (no xfrm_states).*/
2975         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2976         if (IS_ERR(xdst)) {
2977                 xfrm_pols_put(pols, num_pols);
2978                 return ERR_CAST(xdst);
2979         }
2980         xdst->num_pols = num_pols;
2981         xdst->num_xfrms = num_xfrms;
2982         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2983
2984         return xdst;
2985
2986 inc_error:
2987         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2988 error:
2989         xfrm_pols_put(pols, num_pols);
2990         return ERR_PTR(err);
2991 }
2992
2993 static struct dst_entry *make_blackhole(struct net *net, u16 family,
2994                                         struct dst_entry *dst_orig)
2995 {
2996         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2997         struct dst_entry *ret;
2998
2999         if (!afinfo) {
3000                 dst_release(dst_orig);
3001                 return ERR_PTR(-EINVAL);
3002         } else {
3003                 ret = afinfo->blackhole_route(net, dst_orig);
3004         }
3005         rcu_read_unlock();
3006
3007         return ret;
3008 }
3009
3010 /* Finds/creates a bundle for given flow and if_id
3011  *
3012  * At the moment we eat a raw IP route. Mostly to speed up lookups
3013  * on interfaces with disabled IPsec.
3014  *
3015  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3016  * compatibility
3017  */
3018 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3019                                         struct dst_entry *dst_orig,
3020                                         const struct flowi *fl,
3021                                         const struct sock *sk,
3022                                         int flags, u32 if_id)
3023 {
3024         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3025         struct xfrm_dst *xdst;
3026         struct dst_entry *dst, *route;
3027         u16 family = dst_orig->ops->family;
3028         u8 dir = XFRM_POLICY_OUT;
3029         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3030
3031         dst = NULL;
3032         xdst = NULL;
3033         route = NULL;
3034
3035         sk = sk_const_to_full_sk(sk);
3036         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3037                 num_pols = 1;
3038                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3039                                                 if_id);
3040                 err = xfrm_expand_policies(fl, family, pols,
3041                                            &num_pols, &num_xfrms);
3042                 if (err < 0)
3043                         goto dropdst;
3044
3045                 if (num_pols) {
3046                         if (num_xfrms <= 0) {
3047                                 drop_pols = num_pols;
3048                                 goto no_transform;
3049                         }
3050
3051                         xdst = xfrm_resolve_and_create_bundle(
3052                                         pols, num_pols, fl,
3053                                         family, dst_orig);
3054
3055                         if (IS_ERR(xdst)) {
3056                                 xfrm_pols_put(pols, num_pols);
3057                                 err = PTR_ERR(xdst);
3058                                 if (err == -EREMOTE)
3059                                         goto nopol;
3060
3061                                 goto dropdst;
3062                         } else if (xdst == NULL) {
3063                                 num_xfrms = 0;
3064                                 drop_pols = num_pols;
3065                                 goto no_transform;
3066                         }
3067
3068                         route = xdst->route;
3069                 }
3070         }
3071
3072         if (xdst == NULL) {
3073                 struct xfrm_flo xflo;
3074
3075                 xflo.dst_orig = dst_orig;
3076                 xflo.flags = flags;
3077
3078                 /* To accelerate a bit...  */
3079                 if ((dst_orig->flags & DST_NOXFRM) ||
3080                     !net->xfrm.policy_count[XFRM_POLICY_OUT])
3081                         goto nopol;
3082
3083                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3084                 if (xdst == NULL)
3085                         goto nopol;
3086                 if (IS_ERR(xdst)) {
3087                         err = PTR_ERR(xdst);
3088                         goto dropdst;
3089                 }
3090
3091                 num_pols = xdst->num_pols;
3092                 num_xfrms = xdst->num_xfrms;
3093                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3094                 route = xdst->route;
3095         }
3096
3097         dst = &xdst->u.dst;
3098         if (route == NULL && num_xfrms > 0) {
3099                 /* The only case when xfrm_bundle_lookup() returns a
3100                  * bundle with null route, is when the template could
3101                  * not be resolved. It means policies are there, but
3102                  * bundle could not be created, since we don't yet
3103                  * have the xfrm_state's. We need to wait for KM to
3104                  * negotiate new SA's or bail out with error.*/
3105                 if (net->xfrm.sysctl_larval_drop) {
3106                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3107                         err = -EREMOTE;
3108                         goto error;
3109                 }
3110
3111                 err = -EAGAIN;
3112
3113                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3114                 goto error;
3115         }
3116
3117 no_transform:
3118         if (num_pols == 0)
3119                 goto nopol;
3120
3121         if ((flags & XFRM_LOOKUP_ICMP) &&
3122             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3123                 err = -ENOENT;
3124                 goto error;
3125         }
3126
3127         for (i = 0; i < num_pols; i++)
3128                 pols[i]->curlft.use_time = ktime_get_real_seconds();
3129
3130         if (num_xfrms < 0) {
3131                 /* Prohibit the flow */
3132                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3133                 err = -EPERM;
3134                 goto error;
3135         } else if (num_xfrms > 0) {
3136                 /* Flow transformed */
3137                 dst_release(dst_orig);
3138         } else {
3139                 /* Flow passes untransformed */
3140                 dst_release(dst);
3141                 dst = dst_orig;
3142         }
3143 ok:
3144         xfrm_pols_put(pols, drop_pols);
3145         if (dst && dst->xfrm &&
3146             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3147                 dst->flags |= DST_XFRM_TUNNEL;
3148         return dst;
3149
3150 nopol:
3151         if (!(flags & XFRM_LOOKUP_ICMP)) {
3152                 dst = dst_orig;
3153                 goto ok;
3154         }
3155         err = -ENOENT;
3156 error:
3157         dst_release(dst);
3158 dropdst:
3159         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3160                 dst_release(dst_orig);
3161         xfrm_pols_put(pols, drop_pols);
3162         return ERR_PTR(err);
3163 }
3164 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3165
3166 /* Main function: finds/creates a bundle for given flow.
3167  *
3168  * At the moment we eat a raw IP route. Mostly to speed up lookups
3169  * on interfaces with disabled IPsec.
3170  */
3171 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3172                               const struct flowi *fl, const struct sock *sk,
3173                               int flags)
3174 {
3175         return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3176 }
3177 EXPORT_SYMBOL(xfrm_lookup);
3178
3179 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3180  * Otherwise we may send out blackholed packets.
3181  */
3182 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3183                                     const struct flowi *fl,
3184                                     const struct sock *sk, int flags)
3185 {
3186         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3187                                             flags | XFRM_LOOKUP_QUEUE |
3188                                             XFRM_LOOKUP_KEEP_DST_REF);
3189
3190         if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE)
3191                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3192
3193         if (IS_ERR(dst))
3194                 dst_release(dst_orig);
3195
3196         return dst;
3197 }
3198 EXPORT_SYMBOL(xfrm_lookup_route);
3199
3200 static inline int
3201 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3202 {
3203         struct sec_path *sp = skb_sec_path(skb);
3204         struct xfrm_state *x;
3205
3206         if (!sp || idx < 0 || idx >= sp->len)
3207                 return 0;
3208         x = sp->xvec[idx];
3209         if (!x->type->reject)
3210                 return 0;
3211         return x->type->reject(x, skb, fl);
3212 }
3213
3214 /* When skb is transformed back to its "native" form, we have to
3215  * check policy restrictions. At the moment we make this in maximally
3216  * stupid way. Shame on me. :-) Of course, connected sockets must
3217  * have policy cached at them.
3218  */
3219
3220 static inline int
3221 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3222               unsigned short family)
3223 {
3224         if (xfrm_state_kern(x))
3225                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3226         return  x->id.proto == tmpl->id.proto &&
3227                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3228                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3229                 x->props.mode == tmpl->mode &&
3230                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3231                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3232                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3233                   xfrm_state_addr_cmp(tmpl, x, family));
3234 }
3235
3236 /*
3237  * 0 or more than 0 is returned when validation is succeeded (either bypass
3238  * because of optional transport mode, or next index of the mathced secpath
3239  * state with the template.
3240  * -1 is returned when no matching template is found.
3241  * Otherwise "-2 - errored_index" is returned.
3242  */
3243 static inline int
3244 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3245                unsigned short family)
3246 {
3247         int idx = start;
3248
3249         if (tmpl->optional) {
3250                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3251                         return start;
3252         } else
3253                 start = -1;
3254         for (; idx < sp->len; idx++) {
3255                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3256                         return ++idx;
3257                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3258                         if (start == -1)
3259                                 start = -2-idx;
3260                         break;
3261                 }
3262         }
3263         return start;
3264 }
3265
3266 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3267                           unsigned int family, int reverse)
3268 {
3269         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3270         int err;
3271
3272         if (unlikely(afinfo == NULL))
3273                 return -EAFNOSUPPORT;
3274
3275         afinfo->decode_session(skb, fl, reverse);
3276
3277         err = security_xfrm_decode_session(skb, &fl->flowi_secid);
3278         rcu_read_unlock();
3279         return err;
3280 }
3281 EXPORT_SYMBOL(__xfrm_decode_session);
3282
3283 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3284 {
3285         for (; k < sp->len; k++) {
3286                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3287                         *idxp = k;
3288                         return 1;
3289                 }
3290         }
3291
3292         return 0;
3293 }
3294
3295 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3296                         unsigned short family)
3297 {
3298         struct net *net = dev_net(skb->dev);
3299         struct xfrm_policy *pol;
3300         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3301         int npols = 0;
3302         int xfrm_nr;
3303         int pi;
3304         int reverse;
3305         struct flowi fl;
3306         int xerr_idx = -1;
3307         const struct xfrm_if_cb *ifcb;
3308         struct sec_path *sp;
3309         struct xfrm_if *xi;
3310         u32 if_id = 0;
3311
3312         rcu_read_lock();
3313         ifcb = xfrm_if_get_cb();
3314
3315         if (ifcb) {
3316                 xi = ifcb->decode_session(skb);
3317                 if (xi)
3318                         if_id = xi->p.if_id;
3319         }
3320         rcu_read_unlock();
3321
3322         reverse = dir & ~XFRM_POLICY_MASK;
3323         dir &= XFRM_POLICY_MASK;
3324
3325         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3326                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3327                 return 0;
3328         }
3329
3330         nf_nat_decode_session(skb, &fl, family);
3331
3332         /* First, check used SA against their selectors. */
3333         sp = skb_sec_path(skb);
3334         if (sp) {
3335                 int i;
3336
3337                 for (i = sp->len - 1; i >= 0; i--) {
3338                         struct xfrm_state *x = sp->xvec[i];
3339                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
3340                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3341                                 return 0;
3342                         }
3343                 }
3344         }
3345
3346         pol = NULL;
3347         sk = sk_to_full_sk(sk);
3348         if (sk && sk->sk_policy[dir]) {
3349                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3350                 if (IS_ERR(pol)) {
3351                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3352                         return 0;
3353                 }
3354         }
3355
3356         if (!pol)
3357                 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3358
3359         if (IS_ERR(pol)) {
3360                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3361                 return 0;
3362         }
3363
3364         if (!pol) {
3365                 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3366                         xfrm_secpath_reject(xerr_idx, skb, &fl);
3367                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3368                         return 0;
3369                 }
3370                 return 1;
3371         }
3372
3373         pol->curlft.use_time = ktime_get_real_seconds();
3374
3375         pols[0] = pol;
3376         npols++;
3377 #ifdef CONFIG_XFRM_SUB_POLICY
3378         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3379                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3380                                                     &fl, family,
3381                                                     XFRM_POLICY_IN, if_id);
3382                 if (pols[1]) {
3383                         if (IS_ERR(pols[1])) {
3384                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3385                                 return 0;
3386                         }
3387                         pols[1]->curlft.use_time = ktime_get_real_seconds();
3388                         npols++;
3389                 }
3390         }
3391 #endif
3392
3393         if (pol->action == XFRM_POLICY_ALLOW) {
3394                 static struct sec_path dummy;
3395                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3396                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3397                 struct xfrm_tmpl **tpp = tp;
3398                 int ti = 0;
3399                 int i, k;
3400
3401                 sp = skb_sec_path(skb);
3402                 if (!sp)
3403                         sp = &dummy;
3404
3405                 for (pi = 0; pi < npols; pi++) {
3406                         if (pols[pi] != pol &&
3407                             pols[pi]->action != XFRM_POLICY_ALLOW) {
3408                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3409                                 goto reject;
3410                         }
3411                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3412                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3413                                 goto reject_error;
3414                         }
3415                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
3416                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3417                 }
3418                 xfrm_nr = ti;
3419                 if (npols > 1) {
3420                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net);
3421                         tpp = stp;
3422                 }
3423
3424                 /* For each tunnel xfrm, find the first matching tmpl.
3425                  * For each tmpl before that, find corresponding xfrm.
3426                  * Order is _important_. Later we will implement
3427                  * some barriers, but at the moment barriers
3428                  * are implied between each two transformations.
3429                  */
3430                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3431                         k = xfrm_policy_ok(tpp[i], sp, k, family);
3432                         if (k < 0) {
3433                                 if (k < -1)
3434                                         /* "-2 - errored_index" returned */
3435                                         xerr_idx = -(2+k);
3436                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3437                                 goto reject;
3438                         }
3439                 }
3440
3441                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3442                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3443                         goto reject;
3444                 }
3445
3446                 xfrm_pols_put(pols, npols);
3447                 return 1;
3448         }
3449         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3450
3451 reject:
3452         xfrm_secpath_reject(xerr_idx, skb, &fl);
3453 reject_error:
3454         xfrm_pols_put(pols, npols);
3455         return 0;
3456 }
3457 EXPORT_SYMBOL(__xfrm_policy_check);
3458
3459 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3460 {
3461         struct net *net = dev_net(skb->dev);
3462         struct flowi fl;
3463         struct dst_entry *dst;
3464         int res = 1;
3465
3466         if (xfrm_decode_session(skb, &fl, family) < 0) {
3467                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3468                 return 0;
3469         }
3470
3471         skb_dst_force(skb);
3472         if (!skb_dst(skb)) {
3473                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3474                 return 0;
3475         }
3476
3477         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3478         if (IS_ERR(dst)) {
3479                 res = 0;
3480                 dst = NULL;
3481         }
3482         skb_dst_set(skb, dst);
3483         return res;
3484 }
3485 EXPORT_SYMBOL(__xfrm_route_forward);
3486
3487 /* Optimize later using cookies and generation ids. */
3488
3489 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3490 {
3491         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3492          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3493          * get validated by dst_ops->check on every use.  We do this
3494          * because when a normal route referenced by an XFRM dst is
3495          * obsoleted we do not go looking around for all parent
3496          * referencing XFRM dsts so that we can invalidate them.  It
3497          * is just too much work.  Instead we make the checks here on
3498          * every use.  For example:
3499          *
3500          *      XFRM dst A --> IPv4 dst X
3501          *
3502          * X is the "xdst->route" of A (X is also the "dst->path" of A
3503          * in this example).  If X is marked obsolete, "A" will not
3504          * notice.  That's what we are validating here via the
3505          * stale_bundle() check.
3506          *
3507          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3508          * be marked on it.
3509          * This will force stale_bundle() to fail on any xdst bundle with
3510          * this dst linked in it.
3511          */
3512         if (dst->obsolete < 0 && !stale_bundle(dst))
3513                 return dst;
3514
3515         return NULL;
3516 }
3517
3518 static int stale_bundle(struct dst_entry *dst)
3519 {
3520         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3521 }
3522
3523 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3524 {
3525         while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3526                 dst->dev = dev_net(dev)->loopback_dev;
3527                 dev_hold(dst->dev);
3528                 dev_put(dev);
3529         }
3530 }
3531 EXPORT_SYMBOL(xfrm_dst_ifdown);
3532
3533 static void xfrm_link_failure(struct sk_buff *skb)
3534 {
3535         /* Impossible. Such dst must be popped before reaches point of failure. */
3536 }
3537
3538 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3539 {
3540         if (dst) {
3541                 if (dst->obsolete) {
3542                         dst_release(dst);
3543                         dst = NULL;
3544                 }
3545         }
3546         return dst;
3547 }
3548
3549 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3550 {
3551         while (nr--) {
3552                 struct xfrm_dst *xdst = bundle[nr];
3553                 u32 pmtu, route_mtu_cached;
3554                 struct dst_entry *dst;
3555
3556                 dst = &xdst->u.dst;
3557                 pmtu = dst_mtu(xfrm_dst_child(dst));
3558                 xdst->child_mtu_cached = pmtu;
3559
3560                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3561
3562                 route_mtu_cached = dst_mtu(xdst->route);
3563                 xdst->route_mtu_cached = route_mtu_cached;
3564
3565                 if (pmtu > route_mtu_cached)
3566                         pmtu = route_mtu_cached;
3567
3568                 dst_metric_set(dst, RTAX_MTU, pmtu);
3569         }
3570 }
3571
3572 /* Check that the bundle accepts the flow and its components are
3573  * still valid.
3574  */
3575
3576 static int xfrm_bundle_ok(struct xfrm_dst *first)
3577 {
3578         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3579         struct dst_entry *dst = &first->u.dst;
3580         struct xfrm_dst *xdst;
3581         int start_from, nr;
3582         u32 mtu;
3583
3584         if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3585             (dst->dev && !netif_running(dst->dev)))
3586                 return 0;
3587
3588         if (dst->flags & DST_XFRM_QUEUE)
3589                 return 1;
3590
3591         start_from = nr = 0;
3592         do {
3593                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3594
3595                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3596                         return 0;
3597                 if (xdst->xfrm_genid != dst->xfrm->genid)
3598                         return 0;
3599                 if (xdst->num_pols > 0 &&
3600                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3601                         return 0;
3602
3603                 bundle[nr++] = xdst;
3604
3605                 mtu = dst_mtu(xfrm_dst_child(dst));
3606                 if (xdst->child_mtu_cached != mtu) {
3607                         start_from = nr;
3608                         xdst->child_mtu_cached = mtu;
3609                 }
3610
3611                 if (!dst_check(xdst->route, xdst->route_cookie))
3612                         return 0;
3613                 mtu = dst_mtu(xdst->route);
3614                 if (xdst->route_mtu_cached != mtu) {
3615                         start_from = nr;
3616                         xdst->route_mtu_cached = mtu;
3617                 }
3618
3619                 dst = xfrm_dst_child(dst);
3620         } while (dst->xfrm);
3621
3622         if (likely(!start_from))
3623                 return 1;
3624
3625         xdst = bundle[start_from - 1];
3626         mtu = xdst->child_mtu_cached;
3627         while (start_from--) {
3628                 dst = &xdst->u.dst;
3629
3630                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3631                 if (mtu > xdst->route_mtu_cached)
3632                         mtu = xdst->route_mtu_cached;
3633                 dst_metric_set(dst, RTAX_MTU, mtu);
3634                 if (!start_from)
3635                         break;
3636
3637                 xdst = bundle[start_from - 1];
3638                 xdst->child_mtu_cached = mtu;
3639         }
3640
3641         return 1;
3642 }
3643
3644 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3645 {
3646         return dst_metric_advmss(xfrm_dst_path(dst));
3647 }
3648
3649 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3650 {
3651         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3652
3653         return mtu ? : dst_mtu(xfrm_dst_path(dst));
3654 }
3655
3656 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3657                                         const void *daddr)
3658 {
3659         while (dst->xfrm) {
3660                 const struct xfrm_state *xfrm = dst->xfrm;
3661
3662                 dst = xfrm_dst_child(dst);
3663
3664                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3665                         continue;
3666                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3667                         daddr = xfrm->coaddr;
3668                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3669                         daddr = &xfrm->id.daddr;
3670         }
3671         return daddr;
3672 }
3673
3674 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3675                                            struct sk_buff *skb,
3676                                            const void *daddr)
3677 {
3678         const struct dst_entry *path = xfrm_dst_path(dst);
3679
3680         if (!skb)
3681                 daddr = xfrm_get_dst_nexthop(dst, daddr);
3682         return path->ops->neigh_lookup(path, skb, daddr);
3683 }
3684
3685 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3686 {
3687         const struct dst_entry *path = xfrm_dst_path(dst);
3688
3689         daddr = xfrm_get_dst_nexthop(dst, daddr);
3690         path->ops->confirm_neigh(path, daddr);
3691 }
3692
3693 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3694 {
3695         int err = 0;
3696
3697         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3698                 return -EAFNOSUPPORT;
3699
3700         spin_lock(&xfrm_policy_afinfo_lock);
3701         if (unlikely(xfrm_policy_afinfo[family] != NULL))
3702                 err = -EEXIST;
3703         else {
3704                 struct dst_ops *dst_ops = afinfo->dst_ops;
3705                 if (likely(dst_ops->kmem_cachep == NULL))
3706                         dst_ops->kmem_cachep = xfrm_dst_cache;
3707                 if (likely(dst_ops->check == NULL))
3708                         dst_ops->check = xfrm_dst_check;
3709                 if (likely(dst_ops->default_advmss == NULL))
3710                         dst_ops->default_advmss = xfrm_default_advmss;
3711                 if (likely(dst_ops->mtu == NULL))
3712                         dst_ops->mtu = xfrm_mtu;
3713                 if (likely(dst_ops->negative_advice == NULL))
3714                         dst_ops->negative_advice = xfrm_negative_advice;
3715                 if (likely(dst_ops->link_failure == NULL))
3716                         dst_ops->link_failure = xfrm_link_failure;
3717                 if (likely(dst_ops->neigh_lookup == NULL))
3718                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
3719                 if (likely(!dst_ops->confirm_neigh))
3720                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
3721                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3722         }
3723         spin_unlock(&xfrm_policy_afinfo_lock);
3724
3725         return err;
3726 }
3727 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3728
3729 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3730 {
3731         struct dst_ops *dst_ops = afinfo->dst_ops;
3732         int i;
3733
3734         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3735                 if (xfrm_policy_afinfo[i] != afinfo)
3736                         continue;
3737                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3738                 break;
3739         }
3740
3741         synchronize_rcu();
3742
3743         dst_ops->kmem_cachep = NULL;
3744         dst_ops->check = NULL;
3745         dst_ops->negative_advice = NULL;
3746         dst_ops->link_failure = NULL;
3747 }
3748 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3749
3750 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3751 {
3752         spin_lock(&xfrm_if_cb_lock);
3753         rcu_assign_pointer(xfrm_if_cb, ifcb);
3754         spin_unlock(&xfrm_if_cb_lock);
3755 }
3756 EXPORT_SYMBOL(xfrm_if_register_cb);
3757
3758 void xfrm_if_unregister_cb(void)
3759 {
3760         RCU_INIT_POINTER(xfrm_if_cb, NULL);
3761         synchronize_rcu();
3762 }
3763 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3764
3765 #ifdef CONFIG_XFRM_STATISTICS
3766 static int __net_init xfrm_statistics_init(struct net *net)
3767 {
3768         int rv;
3769         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3770         if (!net->mib.xfrm_statistics)
3771                 return -ENOMEM;
3772         rv = xfrm_proc_init(net);
3773         if (rv < 0)
3774                 free_percpu(net->mib.xfrm_statistics);
3775         return rv;
3776 }
3777
3778 static void xfrm_statistics_fini(struct net *net)
3779 {
3780         xfrm_proc_fini(net);
3781         free_percpu(net->mib.xfrm_statistics);
3782 }
3783 #else
3784 static int __net_init xfrm_statistics_init(struct net *net)
3785 {
3786         return 0;
3787 }
3788
3789 static void xfrm_statistics_fini(struct net *net)
3790 {
3791 }
3792 #endif
3793
3794 static int __net_init xfrm_policy_init(struct net *net)
3795 {
3796         unsigned int hmask, sz;
3797         int dir, err;
3798
3799         if (net_eq(net, &init_net)) {
3800                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
3801                                            sizeof(struct xfrm_dst),
3802                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3803                                            NULL);
3804                 err = rhashtable_init(&xfrm_policy_inexact_table,
3805                                       &xfrm_pol_inexact_params);
3806                 BUG_ON(err);
3807         }
3808
3809         hmask = 8 - 1;
3810         sz = (hmask+1) * sizeof(struct hlist_head);
3811
3812         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
3813         if (!net->xfrm.policy_byidx)
3814                 goto out_byidx;
3815         net->xfrm.policy_idx_hmask = hmask;
3816
3817         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
3818                 struct xfrm_policy_hash *htab;
3819
3820                 net->xfrm.policy_count[dir] = 0;
3821                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
3822                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
3823
3824                 htab = &net->xfrm.policy_bydst[dir];
3825                 htab->table = xfrm_hash_alloc(sz);
3826                 if (!htab->table)
3827                         goto out_bydst;
3828                 htab->hmask = hmask;
3829                 htab->dbits4 = 32;
3830                 htab->sbits4 = 32;
3831                 htab->dbits6 = 128;
3832                 htab->sbits6 = 128;
3833         }
3834         net->xfrm.policy_hthresh.lbits4 = 32;
3835         net->xfrm.policy_hthresh.rbits4 = 32;
3836         net->xfrm.policy_hthresh.lbits6 = 128;
3837         net->xfrm.policy_hthresh.rbits6 = 128;
3838
3839         seqlock_init(&net->xfrm.policy_hthresh.lock);
3840
3841         INIT_LIST_HEAD(&net->xfrm.policy_all);
3842         INIT_LIST_HEAD(&net->xfrm.inexact_bins);
3843         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
3844         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
3845         return 0;
3846
3847 out_bydst:
3848         for (dir--; dir >= 0; dir--) {
3849                 struct xfrm_policy_hash *htab;
3850
3851                 htab = &net->xfrm.policy_bydst[dir];
3852                 xfrm_hash_free(htab->table, sz);
3853         }
3854         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3855 out_byidx:
3856         return -ENOMEM;
3857 }
3858
3859 static void xfrm_policy_fini(struct net *net)
3860 {
3861         struct xfrm_pol_inexact_bin *b, *t;
3862         unsigned int sz;
3863         int dir;
3864
3865         flush_work(&net->xfrm.policy_hash_work);
3866 #ifdef CONFIG_XFRM_SUB_POLICY
3867         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
3868 #endif
3869         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
3870
3871         WARN_ON(!list_empty(&net->xfrm.policy_all));
3872
3873         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
3874                 struct xfrm_policy_hash *htab;
3875
3876                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
3877
3878                 htab = &net->xfrm.policy_bydst[dir];
3879                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
3880                 WARN_ON(!hlist_empty(htab->table));
3881                 xfrm_hash_free(htab->table, sz);
3882         }
3883
3884         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
3885         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
3886         xfrm_hash_free(net->xfrm.policy_byidx, sz);
3887
3888         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
3889         list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
3890                 __xfrm_policy_inexact_prune_bin(b, true);
3891         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
3892 }
3893
3894 static int __net_init xfrm_net_init(struct net *net)
3895 {
3896         int rv;
3897
3898         /* Initialize the per-net locks here */
3899         spin_lock_init(&net->xfrm.xfrm_state_lock);
3900         spin_lock_init(&net->xfrm.xfrm_policy_lock);
3901         mutex_init(&net->xfrm.xfrm_cfg_mutex);
3902
3903         rv = xfrm_statistics_init(net);
3904         if (rv < 0)
3905                 goto out_statistics;
3906         rv = xfrm_state_init(net);
3907         if (rv < 0)
3908                 goto out_state;
3909         rv = xfrm_policy_init(net);
3910         if (rv < 0)
3911                 goto out_policy;
3912         rv = xfrm_sysctl_init(net);
3913         if (rv < 0)
3914                 goto out_sysctl;
3915
3916         return 0;
3917
3918 out_sysctl:
3919         xfrm_policy_fini(net);
3920 out_policy:
3921         xfrm_state_fini(net);
3922 out_state:
3923         xfrm_statistics_fini(net);
3924 out_statistics:
3925         return rv;
3926 }
3927
3928 static void __net_exit xfrm_net_exit(struct net *net)
3929 {
3930         xfrm_sysctl_fini(net);
3931         xfrm_policy_fini(net);
3932         xfrm_state_fini(net);
3933         xfrm_statistics_fini(net);
3934 }
3935
3936 static struct pernet_operations __net_initdata xfrm_net_ops = {
3937         .init = xfrm_net_init,
3938         .exit = xfrm_net_exit,
3939 };
3940
3941 void __init xfrm_init(void)
3942 {
3943         register_pernet_subsys(&xfrm_net_ops);
3944         xfrm_dev_init();
3945         seqcount_init(&xfrm_policy_hash_generation);
3946         xfrm_input_init();
3947
3948         RCU_INIT_POINTER(xfrm_if_cb, NULL);
3949         synchronize_rcu();
3950 }
3951
3952 #ifdef CONFIG_AUDITSYSCALL
3953 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
3954                                          struct audit_buffer *audit_buf)
3955 {
3956         struct xfrm_sec_ctx *ctx = xp->security;
3957         struct xfrm_selector *sel = &xp->selector;
3958
3959         if (ctx)
3960                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
3961                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
3962
3963         switch (sel->family) {
3964         case AF_INET:
3965                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
3966                 if (sel->prefixlen_s != 32)
3967                         audit_log_format(audit_buf, " src_prefixlen=%d",
3968                                          sel->prefixlen_s);
3969                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
3970                 if (sel->prefixlen_d != 32)
3971                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3972                                          sel->prefixlen_d);
3973                 break;
3974         case AF_INET6:
3975                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
3976                 if (sel->prefixlen_s != 128)
3977                         audit_log_format(audit_buf, " src_prefixlen=%d",
3978                                          sel->prefixlen_s);
3979                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
3980                 if (sel->prefixlen_d != 128)
3981                         audit_log_format(audit_buf, " dst_prefixlen=%d",
3982                                          sel->prefixlen_d);
3983                 break;
3984         }
3985 }
3986
3987 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
3988 {
3989         struct audit_buffer *audit_buf;
3990
3991         audit_buf = xfrm_audit_start("SPD-add");
3992         if (audit_buf == NULL)
3993                 return;
3994         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
3995         audit_log_format(audit_buf, " res=%u", result);
3996         xfrm_audit_common_policyinfo(xp, audit_buf);
3997         audit_log_end(audit_buf);
3998 }
3999 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4000
4001 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4002                               bool task_valid)
4003 {
4004         struct audit_buffer *audit_buf;
4005
4006         audit_buf = xfrm_audit_start("SPD-delete");
4007         if (audit_buf == NULL)
4008                 return;
4009         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4010         audit_log_format(audit_buf, " res=%u", result);
4011         xfrm_audit_common_policyinfo(xp, audit_buf);
4012         audit_log_end(audit_buf);
4013 }
4014 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4015 #endif
4016
4017 #ifdef CONFIG_XFRM_MIGRATE
4018 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4019                                         const struct xfrm_selector *sel_tgt)
4020 {
4021         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4022                 if (sel_tgt->family == sel_cmp->family &&
4023                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4024                                     sel_cmp->family) &&
4025                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4026                                     sel_cmp->family) &&
4027                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4028                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4029                         return true;
4030                 }
4031         } else {
4032                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4033                         return true;
4034                 }
4035         }
4036         return false;
4037 }
4038
4039 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4040                                                     u8 dir, u8 type, struct net *net)
4041 {
4042         struct xfrm_policy *pol, *ret = NULL;
4043         struct hlist_head *chain;
4044         u32 priority = ~0U;
4045
4046         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4047         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4048         hlist_for_each_entry(pol, chain, bydst) {
4049                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4050                     pol->type == type) {
4051                         ret = pol;
4052                         priority = ret->priority;
4053                         break;
4054                 }
4055         }
4056         chain = &net->xfrm.policy_inexact[dir];
4057         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4058                 if ((pol->priority >= priority) && ret)
4059                         break;
4060
4061                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4062                     pol->type == type) {
4063                         ret = pol;
4064                         break;
4065                 }
4066         }
4067
4068         xfrm_pol_hold(ret);
4069
4070         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4071
4072         return ret;
4073 }
4074
4075 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4076 {
4077         int match = 0;
4078
4079         if (t->mode == m->mode && t->id.proto == m->proto &&
4080             (m->reqid == 0 || t->reqid == m->reqid)) {
4081                 switch (t->mode) {
4082                 case XFRM_MODE_TUNNEL:
4083                 case XFRM_MODE_BEET:
4084                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4085                                             m->old_family) &&
4086                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
4087                                             m->old_family)) {
4088                                 match = 1;
4089                         }
4090                         break;
4091                 case XFRM_MODE_TRANSPORT:
4092                         /* in case of transport mode, template does not store
4093                            any IP addresses, hence we just compare mode and
4094                            protocol */
4095                         match = 1;
4096                         break;
4097                 default:
4098                         break;
4099                 }
4100         }
4101         return match;
4102 }
4103
4104 /* update endpoint address(es) of template(s) */
4105 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4106                                struct xfrm_migrate *m, int num_migrate)
4107 {
4108         struct xfrm_migrate *mp;
4109         int i, j, n = 0;
4110
4111         write_lock_bh(&pol->lock);
4112         if (unlikely(pol->walk.dead)) {
4113                 /* target policy has been deleted */
4114                 write_unlock_bh(&pol->lock);
4115                 return -ENOENT;
4116         }
4117
4118         for (i = 0; i < pol->xfrm_nr; i++) {
4119                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4120                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4121                                 continue;
4122                         n++;
4123                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4124                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4125                                 continue;
4126                         /* update endpoints */
4127                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4128                                sizeof(pol->xfrm_vec[i].id.daddr));
4129                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4130                                sizeof(pol->xfrm_vec[i].saddr));
4131                         pol->xfrm_vec[i].encap_family = mp->new_family;
4132                         /* flush bundles */
4133                         atomic_inc(&pol->genid);
4134                 }
4135         }
4136
4137         write_unlock_bh(&pol->lock);
4138
4139         if (!n)
4140                 return -ENODATA;
4141
4142         return 0;
4143 }
4144
4145 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4146 {
4147         int i, j;
4148
4149         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4150                 return -EINVAL;
4151
4152         for (i = 0; i < num_migrate; i++) {
4153                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4154                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4155                         return -EINVAL;
4156
4157                 /* check if there is any duplicated entry */
4158                 for (j = i + 1; j < num_migrate; j++) {
4159                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4160                                     sizeof(m[i].old_daddr)) &&
4161                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4162                                     sizeof(m[i].old_saddr)) &&
4163                             m[i].proto == m[j].proto &&
4164                             m[i].mode == m[j].mode &&
4165                             m[i].reqid == m[j].reqid &&
4166                             m[i].old_family == m[j].old_family)
4167                                 return -EINVAL;
4168                 }
4169         }
4170
4171         return 0;
4172 }
4173
4174 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4175                  struct xfrm_migrate *m, int num_migrate,
4176                  struct xfrm_kmaddress *k, struct net *net,
4177                  struct xfrm_encap_tmpl *encap)
4178 {
4179         int i, err, nx_cur = 0, nx_new = 0;
4180         struct xfrm_policy *pol = NULL;
4181         struct xfrm_state *x, *xc;
4182         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4183         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4184         struct xfrm_migrate *mp;
4185
4186         /* Stage 0 - sanity checks */
4187         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4188                 goto out;
4189
4190         if (dir >= XFRM_POLICY_MAX) {
4191                 err = -EINVAL;
4192                 goto out;
4193         }
4194
4195         /* Stage 1 - find policy */
4196         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4197                 err = -ENOENT;
4198                 goto out;
4199         }
4200
4201         /* Stage 2 - find and update state(s) */
4202         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4203                 if ((x = xfrm_migrate_state_find(mp, net))) {
4204                         x_cur[nx_cur] = x;
4205                         nx_cur++;
4206                         xc = xfrm_state_migrate(x, mp, encap);
4207                         if (xc) {
4208                                 x_new[nx_new] = xc;
4209                                 nx_new++;
4210                         } else {
4211                                 err = -ENODATA;
4212                                 goto restore_state;
4213                         }
4214                 }
4215         }
4216
4217         /* Stage 3 - update policy */
4218         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4219                 goto restore_state;
4220
4221         /* Stage 4 - delete old state(s) */
4222         if (nx_cur) {
4223                 xfrm_states_put(x_cur, nx_cur);
4224                 xfrm_states_delete(x_cur, nx_cur);
4225         }
4226
4227         /* Stage 5 - announce */
4228         km_migrate(sel, dir, type, m, num_migrate, k, encap);
4229
4230         xfrm_pol_put(pol);
4231
4232         return 0;
4233 out:
4234         return err;
4235
4236 restore_state:
4237         if (pol)
4238                 xfrm_pol_put(pol);
4239         if (nx_cur)
4240                 xfrm_states_put(x_cur, nx_cur);
4241         if (nx_new)
4242                 xfrm_states_delete(x_new, nx_new);
4243
4244         return err;
4245 }
4246 EXPORT_SYMBOL(xfrm_migrate);
4247 #endif