net/xfrm/xfrm_policy.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * xfrm_policy.c
   4  *
   5  * Changes:
   6  *      Mitsuru KANDA @USAGI
   7  *      Kazunori MIYAZAWA @USAGI
   8  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   9  *              IPv6 support
  10  *      Kazunori MIYAZAWA @USAGI
  11  *      YOSHIFUJI Hideaki
  12  *              Split up af-specific portion
  13  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
  14  *
  15  */
  16
  17 #include <linux/err.h>
  18 #include <linux/slab.h>
  19 #include <linux/kmod.h>
  20 #include <linux/list.h>
  21 #include <linux/spinlock.h>
  22 #include <linux/workqueue.h>
  23 #include <linux/notifier.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/netfilter.h>
  26 #include <linux/module.h>
  27 #include <linux/cache.h>
  28 #include <linux/cpu.h>
  29 #include <linux/audit.h>
  30 #include <linux/rhashtable.h>
  31 #include <linux/if_tunnel.h>
  32 #include <net/dst.h>
  33 #include <net/flow.h>
  34 #include <net/xfrm.h>
  35 #include <net/ip.h>
  36 #if IS_ENABLED(CONFIG_IPV6_MIP6)
  37 #include <net/mip6.h>
  38 #endif
  39 #ifdef CONFIG_XFRM_STATISTICS
  40 #include <net/snmp.h>
  41 #endif
  42 #ifdef CONFIG_XFRM_ESPINTCP
  43 #include <net/espintcp.h>
  44 #endif
  45
  46 #include "xfrm_hash.h"
  47
  48 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  49 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  50 #define XFRM_MAX_QUEUE_LEN      100
  51
  52 struct xfrm_flo {
  53         struct dst_entry *dst_orig;
  54         u8 flags;
  55 };
  56
  57 /* prefixes smaller than this are stored in lists, not trees. */
  58 #define INEXACT_PREFIXLEN_IPV4  16
  59 #define INEXACT_PREFIXLEN_IPV6  48
  60
  61 struct xfrm_pol_inexact_node {
  62         struct rb_node node;
  63         union {
  64                 xfrm_address_t addr;
  65                 struct rcu_head rcu;
  66         };
  67         u8 prefixlen;
  68
  69         struct rb_root root;
  70
  71         /* the policies matching this node, can be empty list */
  72         struct hlist_head hhead;
  73 };
  74
  75 /* xfrm inexact policy search tree:
  76  * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
  77  *  |
  78  * +---- root_d: sorted by daddr:prefix
  79  * |                 |
  80  * |        xfrm_pol_inexact_node
  81  * |                 |
  82  * |                 +- root: sorted by saddr/prefix
  83  * |                 |              |
  84  * |                 |         xfrm_pol_inexact_node
  85  * |                 |              |
  86  * |                 |              + root: unused
  87  * |                 |              |
  88  * |                 |              + hhead: saddr:daddr policies
  89  * |                 |
  90  * |                 +- coarse policies and all any:daddr policies
  91  * |
  92  * +---- root_s: sorted by saddr:prefix
  93  * |                 |
  94  * |        xfrm_pol_inexact_node
  95  * |                 |
  96  * |                 + root: unused
  97  * |                 |
  98  * |                 + hhead: saddr:any policies
  99  * |
 100  * +---- coarse policies and all any:any policies
 101  *
 102  * Lookups return four candidate lists:
 103  * 1. any:any list from top-level xfrm_pol_inexact_bin
 104  * 2. any:daddr list from daddr tree
 105  * 3. saddr:daddr list from 2nd level daddr tree
 106  * 4. saddr:any list from saddr tree
 107  *
 108  * This result set then needs to be searched for the policy with
 109  * the lowest priority.  If two results have same prio, youngest one wins.
 110  */
 111
 112 struct xfrm_pol_inexact_key {
 113         possible_net_t net;
 114         u32 if_id;
 115         u16 family;
 116         u8 dir, type;
 117 };
 118
 119 struct xfrm_pol_inexact_bin {
 120         struct xfrm_pol_inexact_key k;
 121         struct rhash_head head;
 122         /* list containing '*:*' policies */
 123         struct hlist_head hhead;
 124
 125         seqcount_spinlock_t count;
 126         /* tree sorted by daddr/prefix */
 127         struct rb_root root_d;
 128
 129         /* tree sorted by saddr/prefix */
 130         struct rb_root root_s;
 131
 132         /* slow path below */
 133         struct list_head inexact_bins;
 134         struct rcu_head rcu;
 135 };
 136
 137 enum xfrm_pol_inexact_candidate_type {
 138         XFRM_POL_CAND_BOTH,
 139         XFRM_POL_CAND_SADDR,
 140         XFRM_POL_CAND_DADDR,
 141         XFRM_POL_CAND_ANY,
 142
 143         XFRM_POL_CAND_MAX,
 144 };
 145
 146 struct xfrm_pol_inexact_candidates {
 147         struct hlist_head *res[XFRM_POL_CAND_MAX];
 148 };
 149
 150 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
 151 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
 152
 153 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
 154 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
 155                                                 __read_mostly;
 156
 157 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
 158 static __read_mostly seqcount_mutex_t xfrm_policy_hash_generation;
 159
 160 static struct rhashtable xfrm_policy_inexact_table;
 161 static const struct rhashtable_params xfrm_pol_inexact_params;
 162
 163 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
 164 static int stale_bundle(struct dst_entry *dst);
 165 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
 166 static void xfrm_policy_queue_process(struct timer_list *t);
 167
 168 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
 169 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
 170                                                 int dir);
 171
 172 static struct xfrm_pol_inexact_bin *
 173 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
 174                            u32 if_id);
 175
 176 static struct xfrm_pol_inexact_bin *
 177 xfrm_policy_inexact_lookup_rcu(struct net *net,
 178                                u8 type, u16 family, u8 dir, u32 if_id);
 179 static struct xfrm_policy *
 180 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
 181                         bool excl);
 182 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
 183                                             struct xfrm_policy *policy);
 184
 185 static bool
 186 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
 187                                     struct xfrm_pol_inexact_bin *b,
 188                                     const xfrm_address_t *saddr,
 189                                     const xfrm_address_t *daddr);
 190
 191 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
 192 {
 193         return refcount_inc_not_zero(&policy->refcnt);
 194 }
 195
 196 static inline bool
 197 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 198 {
 199         const struct flowi4 *fl4 = &fl->u.ip4;
 200
 201         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
 202                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
 203                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
 204                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
 205                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
 206                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
 207 }
 208
 209 static inline bool
 210 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 211 {
 212         const struct flowi6 *fl6 = &fl->u.ip6;
 213
 214         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
 215                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
 216                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
 217                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
 218                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
 219                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
 220 }
 221
 222 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
 223                          unsigned short family)
 224 {
 225         switch (family) {
 226         case AF_INET:
 227                 return __xfrm4_selector_match(sel, fl);
 228         case AF_INET6:
 229                 return __xfrm6_selector_match(sel, fl);
 230         }
 231         return false;
 232 }
 233
 234 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 235 {
 236         const struct xfrm_policy_afinfo *afinfo;
 237
 238         if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
 239                 return NULL;
 240         rcu_read_lock();
 241         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 242         if (unlikely(!afinfo))
 243                 rcu_read_unlock();
 244         return afinfo;
 245 }
 246
 247 /* Called with rcu_read_lock(). */
 248 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
 249 {
 250         return rcu_dereference(xfrm_if_cb);
 251 }
 252
 253 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
 254                                     const xfrm_address_t *saddr,
 255                                     const xfrm_address_t *daddr,
 256                                     int family, u32 mark)
 257 {
 258         const struct xfrm_policy_afinfo *afinfo;
 259         struct dst_entry *dst;
 260
 261         afinfo = xfrm_policy_get_afinfo(family);
 262         if (unlikely(afinfo == NULL))
 263                 return ERR_PTR(-EAFNOSUPPORT);
 264
 265         dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
 266
 267         rcu_read_unlock();
 268
 269         return dst;
 270 }
 271 EXPORT_SYMBOL(__xfrm_dst_lookup);
 272
 273 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
 274                                                 int tos, int oif,
 275                                                 xfrm_address_t *prev_saddr,
 276                                                 xfrm_address_t *prev_daddr,
 277                                                 int family, u32 mark)
 278 {
 279         struct net *net = xs_net(x);
 280         xfrm_address_t *saddr = &x->props.saddr;
 281         xfrm_address_t *daddr = &x->id.daddr;
 282         struct dst_entry *dst;
 283
 284         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 285                 saddr = x->coaddr;
 286                 daddr = prev_daddr;
 287         }
 288         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 289                 saddr = prev_saddr;
 290                 daddr = x->coaddr;
 291         }
 292
 293         dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
 294
 295         if (!IS_ERR(dst)) {
 296                 if (prev_saddr != saddr)
 297                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 298                 if (prev_daddr != daddr)
 299                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 300         }
 301
 302         return dst;
 303 }
 304
 305 static inline unsigned long make_jiffies(long secs)
 306 {
 307         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 308                 return MAX_SCHEDULE_TIMEOUT-1;
 309         else
 310                 return secs*HZ;
 311 }
 312
 313 static void xfrm_policy_timer(struct timer_list *t)
 314 {
 315         struct xfrm_policy *xp = from_timer(xp, t, timer);
 316         time64_t now = ktime_get_real_seconds();
 317         time64_t next = TIME64_MAX;
 318         int warn = 0;
 319         int dir;
 320
 321         read_lock(&xp->lock);
 322
 323         if (unlikely(xp->walk.dead))
 324                 goto out;
 325
 326         dir = xfrm_policy_id2dir(xp->index);
 327
 328         if (xp->lft.hard_add_expires_seconds) {
 329                 time64_t tmo = xp->lft.hard_add_expires_seconds +
 330                         xp->curlft.add_time - now;
 331                 if (tmo <= 0)
 332                         goto expired;
 333                 if (tmo < next)
 334                         next = tmo;
 335         }
 336         if (xp->lft.hard_use_expires_seconds) {
 337                 time64_t tmo = xp->lft.hard_use_expires_seconds +
 338                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 339                 if (tmo <= 0)
 340                         goto expired;
 341                 if (tmo < next)
 342                         next = tmo;
 343         }
 344         if (xp->lft.soft_add_expires_seconds) {
 345                 time64_t tmo = xp->lft.soft_add_expires_seconds +
 346                         xp->curlft.add_time - now;
 347                 if (tmo <= 0) {
 348                         warn = 1;
 349                         tmo = XFRM_KM_TIMEOUT;
 350                 }
 351                 if (tmo < next)
 352                         next = tmo;
 353         }
 354         if (xp->lft.soft_use_expires_seconds) {
 355                 time64_t tmo = xp->lft.soft_use_expires_seconds +
 356                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 357                 if (tmo <= 0) {
 358                         warn = 1;
 359                         tmo = XFRM_KM_TIMEOUT;
 360                 }
 361                 if (tmo < next)
 362                         next = tmo;
 363         }
 364
 365         if (warn)
 366                 km_policy_expired(xp, dir, 0, 0);
 367         if (next != TIME64_MAX &&
 368             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 369                 xfrm_pol_hold(xp);
 370
 371 out:
 372         read_unlock(&xp->lock);
 373         xfrm_pol_put(xp);
 374         return;
 375
 376 expired:
 377         read_unlock(&xp->lock);
 378         if (!xfrm_policy_delete(xp, dir))
 379                 km_policy_expired(xp, dir, 1, 0);
 380         xfrm_pol_put(xp);
 381 }
 382
 383 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 384  * SPD calls.
 385  */
 386
 387 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 388 {
 389         struct xfrm_policy *policy;
 390
 391         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 392
 393         if (policy) {
 394                 write_pnet(&policy->xp_net, net);
 395                 INIT_LIST_HEAD(&policy->walk.all);
 396                 INIT_HLIST_NODE(&policy->bydst_inexact_list);
 397                 INIT_HLIST_NODE(&policy->bydst);
 398                 INIT_HLIST_NODE(&policy->byidx);
 399                 rwlock_init(&policy->lock);
 400                 refcount_set(&policy->refcnt, 1);
 401                 skb_queue_head_init(&policy->polq.hold_queue);
 402                 timer_setup(&policy->timer, xfrm_policy_timer, 0);
 403                 timer_setup(&policy->polq.hold_timer,
 404                             xfrm_policy_queue_process, 0);
 405         }
 406         return policy;
 407 }
 408 EXPORT_SYMBOL(xfrm_policy_alloc);
 409
 410 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
 411 {
 412         struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
 413
 414         security_xfrm_policy_free(policy->security);
 415         kfree(policy);
 416 }
 417
 418 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 419
 420 void xfrm_policy_destroy(struct xfrm_policy *policy)
 421 {
 422         BUG_ON(!policy->walk.dead);
 423
 424         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 425                 BUG();
 426
 427         call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
 428 }
 429 EXPORT_SYMBOL(xfrm_policy_destroy);
 430
 431 /* Rule must be locked. Release descendant resources, announce
 432  * entry dead. The rule must be unlinked from lists to the moment.
 433  */
 434
 435 static void xfrm_policy_kill(struct xfrm_policy *policy)
 436 {
 437         write_lock_bh(&policy->lock);
 438         policy->walk.dead = 1;
 439         write_unlock_bh(&policy->lock);
 440
 441         atomic_inc(&policy->genid);
 442
 443         if (del_timer(&policy->polq.hold_timer))
 444                 xfrm_pol_put(policy);
 445         skb_queue_purge(&policy->polq.hold_queue);
 446
 447         if (del_timer(&policy->timer))
 448                 xfrm_pol_put(policy);
 449
 450         xfrm_pol_put(policy);
 451 }
 452
 453 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 454
 455 static inline unsigned int idx_hash(struct net *net, u32 index)
 456 {
 457         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 458 }
 459
 460 /* calculate policy hash thresholds */
 461 static void __get_hash_thresh(struct net *net,
 462                               unsigned short family, int dir,
 463                               u8 *dbits, u8 *sbits)
 464 {
 465         switch (family) {
 466         case AF_INET:
 467                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 468                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 469                 break;
 470
 471         case AF_INET6:
 472                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 473                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 474                 break;
 475
 476         default:
 477                 *dbits = 0;
 478                 *sbits = 0;
 479         }
 480 }
 481
 482 static struct hlist_head *policy_hash_bysel(struct net *net,
 483                                             const struct xfrm_selector *sel,
 484                                             unsigned short family, int dir)
 485 {
 486         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 487         unsigned int hash;
 488         u8 dbits;
 489         u8 sbits;
 490
 491         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 492         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 493
 494         if (hash == hmask + 1)
 495                 return NULL;
 496
 497         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 498                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 499 }
 500
 501 static struct hlist_head *policy_hash_direct(struct net *net,
 502                                              const xfrm_address_t *daddr,
 503                                              const xfrm_address_t *saddr,
 504                                              unsigned short family, int dir)
 505 {
 506         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 507         unsigned int hash;
 508         u8 dbits;
 509         u8 sbits;
 510
 511         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 512         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 513
 514         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 515                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 516 }
 517
 518 static void xfrm_dst_hash_transfer(struct net *net,
 519                                    struct hlist_head *list,
 520                                    struct hlist_head *ndsttable,
 521                                    unsigned int nhashmask,
 522                                    int dir)
 523 {
 524         struct hlist_node *tmp, *entry0 = NULL;
 525         struct xfrm_policy *pol;
 526         unsigned int h0 = 0;
 527         u8 dbits;
 528         u8 sbits;
 529
 530 redo:
 531         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 532                 unsigned int h;
 533
 534                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 535                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 536                                 pol->family, nhashmask, dbits, sbits);
 537                 if (!entry0) {
 538                         hlist_del_rcu(&pol->bydst);
 539                         hlist_add_head_rcu(&pol->bydst, ndsttable + h);
 540                         h0 = h;
 541                 } else {
 542                         if (h != h0)
 543                                 continue;
 544                         hlist_del_rcu(&pol->bydst);
 545                         hlist_add_behind_rcu(&pol->bydst, entry0);
 546                 }
 547                 entry0 = &pol->bydst;
 548         }
 549         if (!hlist_empty(list)) {
 550                 entry0 = NULL;
 551                 goto redo;
 552         }
 553 }
 554
 555 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 556                                    struct hlist_head *nidxtable,
 557                                    unsigned int nhashmask)
 558 {
 559         struct hlist_node *tmp;
 560         struct xfrm_policy *pol;
 561
 562         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 563                 unsigned int h;
 564
 565                 h = __idx_hash(pol->index, nhashmask);
 566                 hlist_add_head(&pol->byidx, nidxtable+h);
 567         }
 568 }
 569
 570 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 571 {
 572         return ((old_hmask + 1) << 1) - 1;
 573 }
 574
 575 static void xfrm_bydst_resize(struct net *net, int dir)
 576 {
 577         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 578         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 579         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 580         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 581         struct hlist_head *odst;
 582         int i;
 583
 584         if (!ndst)
 585                 return;
 586
 587         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 588         write_seqcount_begin(&xfrm_policy_hash_generation);
 589
 590         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 591                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 592
 593         for (i = hmask; i >= 0; i--)
 594                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 595
 596         rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
 597         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 598
 599         write_seqcount_end(&xfrm_policy_hash_generation);
 600         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 601
 602         synchronize_rcu();
 603
 604         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 605 }
 606
 607 static void xfrm_byidx_resize(struct net *net, int total)
 608 {
 609         unsigned int hmask = net->xfrm.policy_idx_hmask;
 610         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 611         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 612         struct hlist_head *oidx = net->xfrm.policy_byidx;
 613         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 614         int i;
 615
 616         if (!nidx)
 617                 return;
 618
 619         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 620
 621         for (i = hmask; i >= 0; i--)
 622                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 623
 624         net->xfrm.policy_byidx = nidx;
 625         net->xfrm.policy_idx_hmask = nhashmask;
 626
 627         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 628
 629         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 630 }
 631
 632 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 633 {
 634         unsigned int cnt = net->xfrm.policy_count[dir];
 635         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 636
 637         if (total)
 638                 *total += cnt;
 639
 640         if ((hmask + 1) < xfrm_policy_hashmax &&
 641             cnt > hmask)
 642                 return 1;
 643
 644         return 0;
 645 }
 646
 647 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 648 {
 649         unsigned int hmask = net->xfrm.policy_idx_hmask;
 650
 651         if ((hmask + 1) < xfrm_policy_hashmax &&
 652             total > hmask)
 653                 return 1;
 654
 655         return 0;
 656 }
 657
 658 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 659 {
 660         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 661         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 662         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 663         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 664         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 665         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 666         si->spdhcnt = net->xfrm.policy_idx_hmask;
 667         si->spdhmcnt = xfrm_policy_hashmax;
 668 }
 669 EXPORT_SYMBOL(xfrm_spd_getinfo);
 670
 671 static DEFINE_MUTEX(hash_resize_mutex);
 672 static void xfrm_hash_resize(struct work_struct *work)
 673 {
 674         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 675         int dir, total;
 676
 677         mutex_lock(&hash_resize_mutex);
 678
 679         total = 0;
 680         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 681                 if (xfrm_bydst_should_resize(net, dir, &total))
 682                         xfrm_bydst_resize(net, dir);
 683         }
 684         if (xfrm_byidx_should_resize(net, total))
 685                 xfrm_byidx_resize(net, total);
 686
 687         mutex_unlock(&hash_resize_mutex);
 688 }
 689
 690 /* Make sure *pol can be inserted into fastbin.
 691  * Useful to check that later insert requests will be successful
 692  * (provided xfrm_policy_lock is held throughout).
 693  */
 694 static struct xfrm_pol_inexact_bin *
 695 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
 696 {
 697         struct xfrm_pol_inexact_bin *bin, *prev;
 698         struct xfrm_pol_inexact_key k = {
 699                 .family = pol->family,
 700                 .type = pol->type,
 701                 .dir = dir,
 702                 .if_id = pol->if_id,
 703         };
 704         struct net *net = xp_net(pol);
 705
 706         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
 707
 708         write_pnet(&k.net, net);
 709         bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
 710                                      xfrm_pol_inexact_params);
 711         if (bin)
 712                 return bin;
 713
 714         bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
 715         if (!bin)
 716                 return NULL;
 717
 718         bin->k = k;
 719         INIT_HLIST_HEAD(&bin->hhead);
 720         bin->root_d = RB_ROOT;
 721         bin->root_s = RB_ROOT;
 722         seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
 723
 724         prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
 725                                                 &bin->k, &bin->head,
 726                                                 xfrm_pol_inexact_params);
 727         if (!prev) {
 728                 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
 729                 return bin;
 730         }
 731
 732         kfree(bin);
 733
 734         return IS_ERR(prev) ? NULL : prev;
 735 }
 736
 737 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
 738                                                int family, u8 prefixlen)
 739 {
 740         if (xfrm_addr_any(addr, family))
 741                 return true;
 742
 743         if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
 744                 return true;
 745
 746         if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
 747                 return true;
 748
 749         return false;
 750 }
 751
 752 static bool
 753 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
 754 {
 755         const xfrm_address_t *addr;
 756         bool saddr_any, daddr_any;
 757         u8 prefixlen;
 758
 759         addr = &policy->selector.saddr;
 760         prefixlen = policy->selector.prefixlen_s;
 761
 762         saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 763                                                        policy->family,
 764                                                        prefixlen);
 765         addr = &policy->selector.daddr;
 766         prefixlen = policy->selector.prefixlen_d;
 767         daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 768                                                        policy->family,
 769                                                        prefixlen);
 770         return saddr_any && daddr_any;
 771 }
 772
 773 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
 774                                        const xfrm_address_t *addr, u8 prefixlen)
 775 {
 776         node->addr = *addr;
 777         node->prefixlen = prefixlen;
 778 }
 779
 780 static struct xfrm_pol_inexact_node *
 781 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
 782 {
 783         struct xfrm_pol_inexact_node *node;
 784
 785         node = kzalloc(sizeof(*node), GFP_ATOMIC);
 786         if (node)
 787                 xfrm_pol_inexact_node_init(node, addr, prefixlen);
 788
 789         return node;
 790 }
 791
 792 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
 793                                   const xfrm_address_t *b,
 794                                   u8 prefixlen, u16 family)
 795 {
 796         u32 ma, mb, mask;
 797         unsigned int pdw, pbi;
 798         int delta = 0;
 799
 800         switch (family) {
 801         case AF_INET:
 802                 if (prefixlen == 0)
 803                         return 0;
 804                 mask = ~0U << (32 - prefixlen);
 805                 ma = ntohl(a->a4) & mask;
 806                 mb = ntohl(b->a4) & mask;
 807                 if (ma < mb)
 808                         delta = -1;
 809                 else if (ma > mb)
 810                         delta = 1;
 811                 break;
 812         case AF_INET6:
 813                 pdw = prefixlen >> 5;
 814                 pbi = prefixlen & 0x1f;
 815
 816                 if (pdw) {
 817                         delta = memcmp(a->a6, b->a6, pdw << 2);
 818                         if (delta)
 819                                 return delta;
 820                 }
 821                 if (pbi) {
 822                         mask = ~0U << (32 - pbi);
 823                         ma = ntohl(a->a6[pdw]) & mask;
 824                         mb = ntohl(b->a6[pdw]) & mask;
 825                         if (ma < mb)
 826                                 delta = -1;
 827                         else if (ma > mb)
 828                                 delta = 1;
 829                 }
 830                 break;
 831         default:
 832                 break;
 833         }
 834
 835         return delta;
 836 }
 837
 838 static void xfrm_policy_inexact_list_reinsert(struct net *net,
 839                                               struct xfrm_pol_inexact_node *n,
 840                                               u16 family)
 841 {
 842         unsigned int matched_s, matched_d;
 843         struct xfrm_policy *policy, *p;
 844
 845         matched_s = 0;
 846         matched_d = 0;
 847
 848         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 849                 struct hlist_node *newpos = NULL;
 850                 bool matches_s, matches_d;
 851
 852                 if (!policy->bydst_reinsert)
 853                         continue;
 854
 855                 WARN_ON_ONCE(policy->family != family);
 856
 857                 policy->bydst_reinsert = false;
 858                 hlist_for_each_entry(p, &n->hhead, bydst) {
 859                         if (policy->priority > p->priority)
 860                                 newpos = &p->bydst;
 861                         else if (policy->priority == p->priority &&
 862                                  policy->pos > p->pos)
 863                                 newpos = &p->bydst;
 864                         else
 865                                 break;
 866                 }
 867
 868                 if (newpos)
 869                         hlist_add_behind_rcu(&policy->bydst, newpos);
 870                 else
 871                         hlist_add_head_rcu(&policy->bydst, &n->hhead);
 872
 873                 /* paranoia checks follow.
 874                  * Check that the reinserted policy matches at least
 875                  * saddr or daddr for current node prefix.
 876                  *
 877                  * Matching both is fine, matching saddr in one policy
 878                  * (but not daddr) and then matching only daddr in another
 879                  * is a bug.
 880                  */
 881                 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
 882                                                    &n->addr,
 883                                                    n->prefixlen,
 884                                                    family) == 0;
 885                 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
 886                                                    &n->addr,
 887                                                    n->prefixlen,
 888                                                    family) == 0;
 889                 if (matches_s && matches_d)
 890                         continue;
 891
 892                 WARN_ON_ONCE(!matches_s && !matches_d);
 893                 if (matches_s)
 894                         matched_s++;
 895                 if (matches_d)
 896                         matched_d++;
 897                 WARN_ON_ONCE(matched_s && matched_d);
 898         }
 899 }
 900
 901 static void xfrm_policy_inexact_node_reinsert(struct net *net,
 902                                               struct xfrm_pol_inexact_node *n,
 903                                               struct rb_root *new,
 904                                               u16 family)
 905 {
 906         struct xfrm_pol_inexact_node *node;
 907         struct rb_node **p, *parent;
 908
 909         /* we should not have another subtree here */
 910         WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
 911 restart:
 912         parent = NULL;
 913         p = &new->rb_node;
 914         while (*p) {
 915                 u8 prefixlen;
 916                 int delta;
 917
 918                 parent = *p;
 919                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
 920
 921                 prefixlen = min(node->prefixlen, n->prefixlen);
 922
 923                 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
 924                                                prefixlen, family);
 925                 if (delta < 0) {
 926                         p = &parent->rb_left;
 927                 } else if (delta > 0) {
 928                         p = &parent->rb_right;
 929                 } else {
 930                         bool same_prefixlen = node->prefixlen == n->prefixlen;
 931                         struct xfrm_policy *tmp;
 932
 933                         hlist_for_each_entry(tmp, &n->hhead, bydst) {
 934                                 tmp->bydst_reinsert = true;
 935                                 hlist_del_rcu(&tmp->bydst);
 936                         }
 937
 938                         node->prefixlen = prefixlen;
 939
 940                         xfrm_policy_inexact_list_reinsert(net, node, family);
 941
 942                         if (same_prefixlen) {
 943                                 kfree_rcu(n, rcu);
 944                                 return;
 945                         }
 946
 947                         rb_erase(*p, new);
 948                         kfree_rcu(n, rcu);
 949                         n = node;
 950                         goto restart;
 951                 }
 952         }
 953
 954         rb_link_node_rcu(&n->node, parent, p);
 955         rb_insert_color(&n->node, new);
 956 }
 957
 958 /* merge nodes v and n */
 959 static void xfrm_policy_inexact_node_merge(struct net *net,
 960                                            struct xfrm_pol_inexact_node *v,
 961                                            struct xfrm_pol_inexact_node *n,
 962                                            u16 family)
 963 {
 964         struct xfrm_pol_inexact_node *node;
 965         struct xfrm_policy *tmp;
 966         struct rb_node *rnode;
 967
 968         /* To-be-merged node v has a subtree.
 969          *
 970          * Dismantle it and insert its nodes to n->root.
 971          */
 972         while ((rnode = rb_first(&v->root)) != NULL) {
 973                 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
 974                 rb_erase(&node->node, &v->root);
 975                 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
 976                                                   family);
 977         }
 978
 979         hlist_for_each_entry(tmp, &v->hhead, bydst) {
 980                 tmp->bydst_reinsert = true;
 981                 hlist_del_rcu(&tmp->bydst);
 982         }
 983
 984         xfrm_policy_inexact_list_reinsert(net, n, family);
 985 }
 986
 987 static struct xfrm_pol_inexact_node *
 988 xfrm_policy_inexact_insert_node(struct net *net,
 989                                 struct rb_root *root,
 990                                 xfrm_address_t *addr,
 991                                 u16 family, u8 prefixlen, u8 dir)
 992 {
 993         struct xfrm_pol_inexact_node *cached = NULL;
 994         struct rb_node **p, *parent = NULL;
 995         struct xfrm_pol_inexact_node *node;
 996
 997         p = &root->rb_node;
 998         while (*p) {
 999                 int delta;
1000
1001                 parent = *p;
1002                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1003
1004                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1005                                                node->prefixlen,
1006                                                family);
1007                 if (delta == 0 && prefixlen >= node->prefixlen) {
1008                         WARN_ON_ONCE(cached); /* ipsec policies got lost */
1009                         return node;
1010                 }
1011
1012                 if (delta < 0)
1013                         p = &parent->rb_left;
1014                 else
1015                         p = &parent->rb_right;
1016
1017                 if (prefixlen < node->prefixlen) {
1018                         delta = xfrm_policy_addr_delta(addr, &node->addr,
1019                                                        prefixlen,
1020                                                        family);
1021                         if (delta)
1022                                 continue;
1023
1024                         /* This node is a subnet of the new prefix. It needs
1025                          * to be removed and re-inserted with the smaller
1026                          * prefix and all nodes that are now also covered
1027                          * by the reduced prefixlen.
1028                          */
1029                         rb_erase(&node->node, root);
1030
1031                         if (!cached) {
1032                                 xfrm_pol_inexact_node_init(node, addr,
1033                                                            prefixlen);
1034                                 cached = node;
1035                         } else {
1036                                 /* This node also falls within the new
1037                                  * prefixlen. Merge the to-be-reinserted
1038                                  * node and this one.
1039                                  */
1040                                 xfrm_policy_inexact_node_merge(net, node,
1041                                                                cached, family);
1042                                 kfree_rcu(node, rcu);
1043                         }
1044
1045                         /* restart */
1046                         p = &root->rb_node;
1047                         parent = NULL;
1048                 }
1049         }
1050
1051         node = cached;
1052         if (!node) {
1053                 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1054                 if (!node)
1055                         return NULL;
1056         }
1057
1058         rb_link_node_rcu(&node->node, parent, p);
1059         rb_insert_color(&node->node, root);
1060
1061         return node;
1062 }
1063
1064 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1065 {
1066         struct xfrm_pol_inexact_node *node;
1067         struct rb_node *rn = rb_first(r);
1068
1069         while (rn) {
1070                 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1071
1072                 xfrm_policy_inexact_gc_tree(&node->root, rm);
1073                 rn = rb_next(rn);
1074
1075                 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1076                         WARN_ON_ONCE(rm);
1077                         continue;
1078                 }
1079
1080                 rb_erase(&node->node, r);
1081                 kfree_rcu(node, rcu);
1082         }
1083 }
1084
1085 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1086 {
1087         write_seqcount_begin(&b->count);
1088         xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1089         xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1090         write_seqcount_end(&b->count);
1091
1092         if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1093             !hlist_empty(&b->hhead)) {
1094                 WARN_ON_ONCE(net_exit);
1095                 return;
1096         }
1097
1098         if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1099                                    xfrm_pol_inexact_params) == 0) {
1100                 list_del(&b->inexact_bins);
1101                 kfree_rcu(b, rcu);
1102         }
1103 }
1104
1105 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1106 {
1107         struct net *net = read_pnet(&b->k.net);
1108
1109         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1110         __xfrm_policy_inexact_prune_bin(b, false);
1111         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1112 }
1113
1114 static void __xfrm_policy_inexact_flush(struct net *net)
1115 {
1116         struct xfrm_pol_inexact_bin *bin, *t;
1117
1118         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1119
1120         list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1121                 __xfrm_policy_inexact_prune_bin(bin, false);
1122 }
1123
1124 static struct hlist_head *
1125 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1126                                 struct xfrm_policy *policy, u8 dir)
1127 {
1128         struct xfrm_pol_inexact_node *n;
1129         struct net *net;
1130
1131         net = xp_net(policy);
1132         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1133
1134         if (xfrm_policy_inexact_insert_use_any_list(policy))
1135                 return &bin->hhead;
1136
1137         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1138                                                policy->family,
1139                                                policy->selector.prefixlen_d)) {
1140                 write_seqcount_begin(&bin->count);
1141                 n = xfrm_policy_inexact_insert_node(net,
1142                                                     &bin->root_s,
1143                                                     &policy->selector.saddr,
1144                                                     policy->family,
1145                                                     policy->selector.prefixlen_s,
1146                                                     dir);
1147                 write_seqcount_end(&bin->count);
1148                 if (!n)
1149                         return NULL;
1150
1151                 return &n->hhead;
1152         }
1153
1154         /* daddr is fixed */
1155         write_seqcount_begin(&bin->count);
1156         n = xfrm_policy_inexact_insert_node(net,
1157                                             &bin->root_d,
1158                                             &policy->selector.daddr,
1159                                             policy->family,
1160                                             policy->selector.prefixlen_d, dir);
1161         write_seqcount_end(&bin->count);
1162         if (!n)
1163                 return NULL;
1164
1165         /* saddr is wildcard */
1166         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1167                                                policy->family,
1168                                                policy->selector.prefixlen_s))
1169                 return &n->hhead;
1170
1171         write_seqcount_begin(&bin->count);
1172         n = xfrm_policy_inexact_insert_node(net,
1173                                             &n->root,
1174                                             &policy->selector.saddr,
1175                                             policy->family,
1176                                             policy->selector.prefixlen_s, dir);
1177         write_seqcount_end(&bin->count);
1178         if (!n)
1179                 return NULL;
1180
1181         return &n->hhead;
1182 }
1183
1184 static struct xfrm_policy *
1185 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1186 {
1187         struct xfrm_pol_inexact_bin *bin;
1188         struct xfrm_policy *delpol;
1189         struct hlist_head *chain;
1190         struct net *net;
1191
1192         bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1193         if (!bin)
1194                 return ERR_PTR(-ENOMEM);
1195
1196         net = xp_net(policy);
1197         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1198
1199         chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1200         if (!chain) {
1201                 __xfrm_policy_inexact_prune_bin(bin, false);
1202                 return ERR_PTR(-ENOMEM);
1203         }
1204
1205         delpol = xfrm_policy_insert_list(chain, policy, excl);
1206         if (delpol && excl) {
1207                 __xfrm_policy_inexact_prune_bin(bin, false);
1208                 return ERR_PTR(-EEXIST);
1209         }
1210
1211         chain = &net->xfrm.policy_inexact[dir];
1212         xfrm_policy_insert_inexact_list(chain, policy);
1213
1214         if (delpol)
1215                 __xfrm_policy_inexact_prune_bin(bin, false);
1216
1217         return delpol;
1218 }
1219
1220 static void xfrm_hash_rebuild(struct work_struct *work)
1221 {
1222         struct net *net = container_of(work, struct net,
1223                                        xfrm.policy_hthresh.work);
1224         unsigned int hmask;
1225         struct xfrm_policy *pol;
1226         struct xfrm_policy *policy;
1227         struct hlist_head *chain;
1228         struct hlist_head *odst;
1229         struct hlist_node *newpos;
1230         int i;
1231         int dir;
1232         unsigned seq;
1233         u8 lbits4, rbits4, lbits6, rbits6;
1234
1235         mutex_lock(&hash_resize_mutex);
1236
1237         /* read selector prefixlen thresholds */
1238         do {
1239                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1240
1241                 lbits4 = net->xfrm.policy_hthresh.lbits4;
1242                 rbits4 = net->xfrm.policy_hthresh.rbits4;
1243                 lbits6 = net->xfrm.policy_hthresh.lbits6;
1244                 rbits6 = net->xfrm.policy_hthresh.rbits6;
1245         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1246
1247         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1248         write_seqcount_begin(&xfrm_policy_hash_generation);
1249
1250         /* make sure that we can insert the indirect policies again before
1251          * we start with destructive action.
1252          */
1253         list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1254                 struct xfrm_pol_inexact_bin *bin;
1255                 u8 dbits, sbits;
1256
1257                 dir = xfrm_policy_id2dir(policy->index);
1258                 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1259                         continue;
1260
1261                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1262                         if (policy->family == AF_INET) {
1263                                 dbits = rbits4;
1264                                 sbits = lbits4;
1265                         } else {
1266                                 dbits = rbits6;
1267                                 sbits = lbits6;
1268                         }
1269                 } else {
1270                         if (policy->family == AF_INET) {
1271                                 dbits = lbits4;
1272                                 sbits = rbits4;
1273                         } else {
1274                                 dbits = lbits6;
1275                                 sbits = rbits6;
1276                         }
1277                 }
1278
1279                 if (policy->selector.prefixlen_d < dbits ||
1280                     policy->selector.prefixlen_s < sbits)
1281                         continue;
1282
1283                 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1284                 if (!bin)
1285                         goto out_unlock;
1286
1287                 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1288                         goto out_unlock;
1289         }
1290
1291         /* reset the bydst and inexact table in all directions */
1292         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1293                 struct hlist_node *n;
1294
1295                 hlist_for_each_entry_safe(policy, n,
1296                                           &net->xfrm.policy_inexact[dir],
1297                                           bydst_inexact_list) {
1298                         hlist_del_rcu(&policy->bydst);
1299                         hlist_del_init(&policy->bydst_inexact_list);
1300                 }
1301
1302                 hmask = net->xfrm.policy_bydst[dir].hmask;
1303                 odst = net->xfrm.policy_bydst[dir].table;
1304                 for (i = hmask; i >= 0; i--) {
1305                         hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1306                                 hlist_del_rcu(&policy->bydst);
1307                 }
1308                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1309                         /* dir out => dst = remote, src = local */
1310                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1311                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1312                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1313                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1314                 } else {
1315                         /* dir in/fwd => dst = local, src = remote */
1316                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1317                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1318                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1319                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1320                 }
1321         }
1322
1323         /* re-insert all policies by order of creation */
1324         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1325                 if (policy->walk.dead)
1326                         continue;
1327                 dir = xfrm_policy_id2dir(policy->index);
1328                 if (dir >= XFRM_POLICY_MAX) {
1329                         /* skip socket policies */
1330                         continue;
1331                 }
1332                 newpos = NULL;
1333                 chain = policy_hash_bysel(net, &policy->selector,
1334                                           policy->family, dir);
1335
1336                 if (!chain) {
1337                         void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1338
1339                         WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1340                         continue;
1341                 }
1342
1343                 hlist_for_each_entry(pol, chain, bydst) {
1344                         if (policy->priority >= pol->priority)
1345                                 newpos = &pol->bydst;
1346                         else
1347                                 break;
1348                 }
1349                 if (newpos)
1350                         hlist_add_behind_rcu(&policy->bydst, newpos);
1351                 else
1352                         hlist_add_head_rcu(&policy->bydst, chain);
1353         }
1354
1355 out_unlock:
1356         __xfrm_policy_inexact_flush(net);
1357         write_seqcount_end(&xfrm_policy_hash_generation);
1358         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1359
1360         mutex_unlock(&hash_resize_mutex);
1361 }
1362
1363 void xfrm_policy_hash_rebuild(struct net *net)
1364 {
1365         schedule_work(&net->xfrm.policy_hthresh.work);
1366 }
1367 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1368
1369 /* Generate new index... KAME seems to generate them ordered by cost
1370  * of an absolute inpredictability of ordering of rules. This will not pass. */
1371 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1372 {
1373         static u32 idx_generator;
1374
1375         for (;;) {
1376                 struct hlist_head *list;
1377                 struct xfrm_policy *p;
1378                 u32 idx;
1379                 int found;
1380
1381                 if (!index) {
1382                         idx = (idx_generator | dir);
1383                         idx_generator += 8;
1384                 } else {
1385                         idx = index;
1386                         index = 0;
1387                 }
1388
1389                 if (idx == 0)
1390                         idx = 8;
1391                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1392                 found = 0;
1393                 hlist_for_each_entry(p, list, byidx) {
1394                         if (p->index == idx) {
1395                                 found = 1;
1396                                 break;
1397                         }
1398                 }
1399                 if (!found)
1400                         return idx;
1401         }
1402 }
1403
1404 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1405 {
1406         u32 *p1 = (u32 *) s1;
1407         u32 *p2 = (u32 *) s2;
1408         int len = sizeof(struct xfrm_selector) / sizeof(u32);
1409         int i;
1410
1411         for (i = 0; i < len; i++) {
1412                 if (p1[i] != p2[i])
1413                         return 1;
1414         }
1415
1416         return 0;
1417 }
1418
1419 static void xfrm_policy_requeue(struct xfrm_policy *old,
1420                                 struct xfrm_policy *new)
1421 {
1422         struct xfrm_policy_queue *pq = &old->polq;
1423         struct sk_buff_head list;
1424
1425         if (skb_queue_empty(&pq->hold_queue))
1426                 return;
1427
1428         __skb_queue_head_init(&list);
1429
1430         spin_lock_bh(&pq->hold_queue.lock);
1431         skb_queue_splice_init(&pq->hold_queue, &list);
1432         if (del_timer(&pq->hold_timer))
1433                 xfrm_pol_put(old);
1434         spin_unlock_bh(&pq->hold_queue.lock);
1435
1436         pq = &new->polq;
1437
1438         spin_lock_bh(&pq->hold_queue.lock);
1439         skb_queue_splice(&list, &pq->hold_queue);
1440         pq->timeout = XFRM_QUEUE_TMO_MIN;
1441         if (!mod_timer(&pq->hold_timer, jiffies))
1442                 xfrm_pol_hold(new);
1443         spin_unlock_bh(&pq->hold_queue.lock);
1444 }
1445
1446 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1447                                           struct xfrm_policy *pol)
1448 {
1449         return mark->v == pol->mark.v && mark->m == pol->mark.m;
1450 }
1451
1452 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1453 {
1454         const struct xfrm_pol_inexact_key *k = data;
1455         u32 a = k->type << 24 | k->dir << 16 | k->family;
1456
1457         return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1458                             seed);
1459 }
1460
1461 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1462 {
1463         const struct xfrm_pol_inexact_bin *b = data;
1464
1465         return xfrm_pol_bin_key(&b->k, 0, seed);
1466 }
1467
1468 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1469                             const void *ptr)
1470 {
1471         const struct xfrm_pol_inexact_key *key = arg->key;
1472         const struct xfrm_pol_inexact_bin *b = ptr;
1473         int ret;
1474
1475         if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1476                 return -1;
1477
1478         ret = b->k.dir ^ key->dir;
1479         if (ret)
1480                 return ret;
1481
1482         ret = b->k.type ^ key->type;
1483         if (ret)
1484                 return ret;
1485
1486         ret = b->k.family ^ key->family;
1487         if (ret)
1488                 return ret;
1489
1490         return b->k.if_id ^ key->if_id;
1491 }
1492
1493 static const struct rhashtable_params xfrm_pol_inexact_params = {
1494         .head_offset            = offsetof(struct xfrm_pol_inexact_bin, head),
1495         .hashfn                 = xfrm_pol_bin_key,
1496         .obj_hashfn             = xfrm_pol_bin_obj,
1497         .obj_cmpfn              = xfrm_pol_bin_cmp,
1498         .automatic_shrinking    = true,
1499 };
1500
1501 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1502                                             struct xfrm_policy *policy)
1503 {
1504         struct xfrm_policy *pol, *delpol = NULL;
1505         struct hlist_node *newpos = NULL;
1506         int i = 0;
1507
1508         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1509                 if (pol->type == policy->type &&
1510                     pol->if_id == policy->if_id &&
1511                     !selector_cmp(&pol->selector, &policy->selector) &&
1512                     xfrm_policy_mark_match(&policy->mark, pol) &&
1513                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1514                     !WARN_ON(delpol)) {
1515                         delpol = pol;
1516                         if (policy->priority > pol->priority)
1517                                 continue;
1518                 } else if (policy->priority >= pol->priority) {
1519                         newpos = &pol->bydst_inexact_list;
1520                         continue;
1521                 }
1522                 if (delpol)
1523                         break;
1524         }
1525
1526         if (newpos)
1527                 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1528         else
1529                 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1530
1531         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1532                 pol->pos = i;
1533                 i++;
1534         }
1535 }
1536
1537 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1538                                                    struct xfrm_policy *policy,
1539                                                    bool excl)
1540 {
1541         struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1542
1543         hlist_for_each_entry(pol, chain, bydst) {
1544                 if (pol->type == policy->type &&
1545                     pol->if_id == policy->if_id &&
1546                     !selector_cmp(&pol->selector, &policy->selector) &&
1547                     xfrm_policy_mark_match(&policy->mark, pol) &&
1548                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1549                     !WARN_ON(delpol)) {
1550                         if (excl)
1551                                 return ERR_PTR(-EEXIST);
1552                         delpol = pol;
1553                         if (policy->priority > pol->priority)
1554                                 continue;
1555                 } else if (policy->priority >= pol->priority) {
1556                         newpos = pol;
1557                         continue;
1558                 }
1559                 if (delpol)
1560                         break;
1561         }
1562
1563         if (newpos)
1564                 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1565         else
1566                 hlist_add_head_rcu(&policy->bydst, chain);
1567
1568         return delpol;
1569 }
1570
1571 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1572 {
1573         struct net *net = xp_net(policy);
1574         struct xfrm_policy *delpol;
1575         struct hlist_head *chain;
1576
1577         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1578         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1579         if (chain)
1580                 delpol = xfrm_policy_insert_list(chain, policy, excl);
1581         else
1582                 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1583
1584         if (IS_ERR(delpol)) {
1585                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1586                 return PTR_ERR(delpol);
1587         }
1588
1589         __xfrm_policy_link(policy, dir);
1590
1591         /* After previous checking, family can either be AF_INET or AF_INET6 */
1592         if (policy->family == AF_INET)
1593                 rt_genid_bump_ipv4(net);
1594         else
1595                 rt_genid_bump_ipv6(net);
1596
1597         if (delpol) {
1598                 xfrm_policy_requeue(delpol, policy);
1599                 __xfrm_policy_unlink(delpol, dir);
1600         }
1601         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1602         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1603         policy->curlft.add_time = ktime_get_real_seconds();
1604         policy->curlft.use_time = 0;
1605         if (!mod_timer(&policy->timer, jiffies + HZ))
1606                 xfrm_pol_hold(policy);
1607         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1608
1609         if (delpol)
1610                 xfrm_policy_kill(delpol);
1611         else if (xfrm_bydst_should_resize(net, dir, NULL))
1612                 schedule_work(&net->xfrm.policy_hash_work);
1613
1614         return 0;
1615 }
1616 EXPORT_SYMBOL(xfrm_policy_insert);
1617
1618 static struct xfrm_policy *
1619 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1620                         u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1621                         struct xfrm_sec_ctx *ctx)
1622 {
1623         struct xfrm_policy *pol;
1624
1625         if (!chain)
1626                 return NULL;
1627
1628         hlist_for_each_entry(pol, chain, bydst) {
1629                 if (pol->type == type &&
1630                     pol->if_id == if_id &&
1631                     xfrm_policy_mark_match(mark, pol) &&
1632                     !selector_cmp(sel, &pol->selector) &&
1633                     xfrm_sec_ctx_match(ctx, pol->security))
1634                         return pol;
1635         }
1636
1637         return NULL;
1638 }
1639
1640 struct xfrm_policy *
1641 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1642                       u8 type, int dir, struct xfrm_selector *sel,
1643                       struct xfrm_sec_ctx *ctx, int delete, int *err)
1644 {
1645         struct xfrm_pol_inexact_bin *bin = NULL;
1646         struct xfrm_policy *pol, *ret = NULL;
1647         struct hlist_head *chain;
1648
1649         *err = 0;
1650         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1651         chain = policy_hash_bysel(net, sel, sel->family, dir);
1652         if (!chain) {
1653                 struct xfrm_pol_inexact_candidates cand;
1654                 int i;
1655
1656                 bin = xfrm_policy_inexact_lookup(net, type,
1657                                                  sel->family, dir, if_id);
1658                 if (!bin) {
1659                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1660                         return NULL;
1661                 }
1662
1663                 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1664                                                          &sel->saddr,
1665                                                          &sel->daddr)) {
1666                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1667                         return NULL;
1668                 }
1669
1670                 pol = NULL;
1671                 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1672                         struct xfrm_policy *tmp;
1673
1674                         tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1675                                                       if_id, type, dir,
1676                                                       sel, ctx);
1677                         if (!tmp)
1678                                 continue;
1679
1680                         if (!pol || tmp->pos < pol->pos)
1681                                 pol = tmp;
1682                 }
1683         } else {
1684                 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1685                                               sel, ctx);
1686         }
1687
1688         if (pol) {
1689                 xfrm_pol_hold(pol);
1690                 if (delete) {
1691                         *err = security_xfrm_policy_delete(pol->security);
1692                         if (*err) {
1693                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1694                                 return pol;
1695                         }
1696                         __xfrm_policy_unlink(pol, dir);
1697                 }
1698                 ret = pol;
1699         }
1700         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1701
1702         if (ret && delete)
1703                 xfrm_policy_kill(ret);
1704         if (bin && delete)
1705                 xfrm_policy_inexact_prune_bin(bin);
1706         return ret;
1707 }
1708 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1709
1710 struct xfrm_policy *
1711 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1712                  u8 type, int dir, u32 id, int delete, int *err)
1713 {
1714         struct xfrm_policy *pol, *ret;
1715         struct hlist_head *chain;
1716
1717         *err = -ENOENT;
1718         if (xfrm_policy_id2dir(id) != dir)
1719                 return NULL;
1720
1721         *err = 0;
1722         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1723         chain = net->xfrm.policy_byidx + idx_hash(net, id);
1724         ret = NULL;
1725         hlist_for_each_entry(pol, chain, byidx) {
1726                 if (pol->type == type && pol->index == id &&
1727                     pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1728                         xfrm_pol_hold(pol);
1729                         if (delete) {
1730                                 *err = security_xfrm_policy_delete(
1731                                                                 pol->security);
1732                                 if (*err) {
1733                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1734                                         return pol;
1735                                 }
1736                                 __xfrm_policy_unlink(pol, dir);
1737                         }
1738                         ret = pol;
1739                         break;
1740                 }
1741         }
1742         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1743
1744         if (ret && delete)
1745                 xfrm_policy_kill(ret);
1746         return ret;
1747 }
1748 EXPORT_SYMBOL(xfrm_policy_byid);
1749
1750 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1751 static inline int
1752 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1753 {
1754         struct xfrm_policy *pol;
1755         int err = 0;
1756
1757         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1758                 if (pol->walk.dead ||
1759                     xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1760                     pol->type != type)
1761                         continue;
1762
1763                 err = security_xfrm_policy_delete(pol->security);
1764                 if (err) {
1765                         xfrm_audit_policy_delete(pol, 0, task_valid);
1766                         return err;
1767                 }
1768         }
1769         return err;
1770 }
1771 #else
1772 static inline int
1773 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1774 {
1775         return 0;
1776 }
1777 #endif
1778
1779 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1780 {
1781         int dir, err = 0, cnt = 0;
1782         struct xfrm_policy *pol;
1783
1784         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1785
1786         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1787         if (err)
1788                 goto out;
1789
1790 again:
1791         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1792                 dir = xfrm_policy_id2dir(pol->index);
1793                 if (pol->walk.dead ||
1794                     dir >= XFRM_POLICY_MAX ||
1795                     pol->type != type)
1796                         continue;
1797
1798                 __xfrm_policy_unlink(pol, dir);
1799                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1800                 cnt++;
1801                 xfrm_audit_policy_delete(pol, 1, task_valid);
1802                 xfrm_policy_kill(pol);
1803                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1804                 goto again;
1805         }
1806         if (cnt)
1807                 __xfrm_policy_inexact_flush(net);
1808         else
1809                 err = -ESRCH;
1810 out:
1811         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1812         return err;
1813 }
1814 EXPORT_SYMBOL(xfrm_policy_flush);
1815
1816 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1817                      int (*func)(struct xfrm_policy *, int, int, void*),
1818                      void *data)
1819 {
1820         struct xfrm_policy *pol;
1821         struct xfrm_policy_walk_entry *x;
1822         int error = 0;
1823
1824         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1825             walk->type != XFRM_POLICY_TYPE_ANY)
1826                 return -EINVAL;
1827
1828         if (list_empty(&walk->walk.all) && walk->seq != 0)
1829                 return 0;
1830
1831         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1832         if (list_empty(&walk->walk.all))
1833                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1834         else
1835                 x = list_first_entry(&walk->walk.all,
1836                                      struct xfrm_policy_walk_entry, all);
1837
1838         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1839                 if (x->dead)
1840                         continue;
1841                 pol = container_of(x, struct xfrm_policy, walk);
1842                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1843                     walk->type != pol->type)
1844                         continue;
1845                 error = func(pol, xfrm_policy_id2dir(pol->index),
1846                              walk->seq, data);
1847                 if (error) {
1848                         list_move_tail(&walk->walk.all, &x->all);
1849                         goto out;
1850                 }
1851                 walk->seq++;
1852         }
1853         if (walk->seq == 0) {
1854                 error = -ENOENT;
1855                 goto out;
1856         }
1857         list_del_init(&walk->walk.all);
1858 out:
1859         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1860         return error;
1861 }
1862 EXPORT_SYMBOL(xfrm_policy_walk);
1863
1864 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1865 {
1866         INIT_LIST_HEAD(&walk->walk.all);
1867         walk->walk.dead = 1;
1868         walk->type = type;
1869         walk->seq = 0;
1870 }
1871 EXPORT_SYMBOL(xfrm_policy_walk_init);
1872
1873 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1874 {
1875         if (list_empty(&walk->walk.all))
1876                 return;
1877
1878         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1879         list_del(&walk->walk.all);
1880         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1881 }
1882 EXPORT_SYMBOL(xfrm_policy_walk_done);
1883
1884 /*
1885  * Find policy to apply to this flow.
1886  *
1887  * Returns 0 if policy found, else an -errno.
1888  */
1889 static int xfrm_policy_match(const struct xfrm_policy *pol,
1890                              const struct flowi *fl,
1891                              u8 type, u16 family, int dir, u32 if_id)
1892 {
1893         const struct xfrm_selector *sel = &pol->selector;
1894         int ret = -ESRCH;
1895         bool match;
1896
1897         if (pol->family != family ||
1898             pol->if_id != if_id ||
1899             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1900             pol->type != type)
1901                 return ret;
1902
1903         match = xfrm_selector_match(sel, fl, family);
1904         if (match)
1905                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid,
1906                                                   dir);
1907         return ret;
1908 }
1909
1910 static struct xfrm_pol_inexact_node *
1911 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1912                                 seqcount_spinlock_t *count,
1913                                 const xfrm_address_t *addr, u16 family)
1914 {
1915         const struct rb_node *parent;
1916         int seq;
1917
1918 again:
1919         seq = read_seqcount_begin(count);
1920
1921         parent = rcu_dereference_raw(r->rb_node);
1922         while (parent) {
1923                 struct xfrm_pol_inexact_node *node;
1924                 int delta;
1925
1926                 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1927
1928                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1929                                                node->prefixlen, family);
1930                 if (delta < 0) {
1931                         parent = rcu_dereference_raw(parent->rb_left);
1932                         continue;
1933                 } else if (delta > 0) {
1934                         parent = rcu_dereference_raw(parent->rb_right);
1935                         continue;
1936                 }
1937
1938                 return node;
1939         }
1940
1941         if (read_seqcount_retry(count, seq))
1942                 goto again;
1943
1944         return NULL;
1945 }
1946
1947 static bool
1948 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1949                                     struct xfrm_pol_inexact_bin *b,
1950                                     const xfrm_address_t *saddr,
1951                                     const xfrm_address_t *daddr)
1952 {
1953         struct xfrm_pol_inexact_node *n;
1954         u16 family;
1955
1956         if (!b)
1957                 return false;
1958
1959         family = b->k.family;
1960         memset(cand, 0, sizeof(*cand));
1961         cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1962
1963         n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1964                                             family);
1965         if (n) {
1966                 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1967                 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1968                                                     family);
1969                 if (n)
1970                         cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1971         }
1972
1973         n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1974                                             family);
1975         if (n)
1976                 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1977
1978         return true;
1979 }
1980
1981 static struct xfrm_pol_inexact_bin *
1982 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1983                                u8 dir, u32 if_id)
1984 {
1985         struct xfrm_pol_inexact_key k = {
1986                 .family = family,
1987                 .type = type,
1988                 .dir = dir,
1989                 .if_id = if_id,
1990         };
1991
1992         write_pnet(&k.net, net);
1993
1994         return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1995                                  xfrm_pol_inexact_params);
1996 }
1997
1998 static struct xfrm_pol_inexact_bin *
1999 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2000                            u8 dir, u32 if_id)
2001 {
2002         struct xfrm_pol_inexact_bin *bin;
2003
2004         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2005
2006         rcu_read_lock();
2007         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2008         rcu_read_unlock();
2009
2010         return bin;
2011 }
2012
2013 static struct xfrm_policy *
2014 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2015                               struct xfrm_policy *prefer,
2016                               const struct flowi *fl,
2017                               u8 type, u16 family, int dir, u32 if_id)
2018 {
2019         u32 priority = prefer ? prefer->priority : ~0u;
2020         struct xfrm_policy *pol;
2021
2022         if (!chain)
2023                 return NULL;
2024
2025         hlist_for_each_entry_rcu(pol, chain, bydst) {
2026                 int err;
2027
2028                 if (pol->priority > priority)
2029                         break;
2030
2031                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2032                 if (err) {
2033                         if (err != -ESRCH)
2034                                 return ERR_PTR(err);
2035
2036                         continue;
2037                 }
2038
2039                 if (prefer) {
2040                         /* matches.  Is it older than *prefer? */
2041                         if (pol->priority == priority &&
2042                             prefer->pos < pol->pos)
2043                                 return prefer;
2044                 }
2045
2046                 return pol;
2047         }
2048
2049         return NULL;
2050 }
2051
2052 static struct xfrm_policy *
2053 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2054                             struct xfrm_policy *prefer,
2055                             const struct flowi *fl,
2056                             u8 type, u16 family, int dir, u32 if_id)
2057 {
2058         struct xfrm_policy *tmp;
2059         int i;
2060
2061         for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2062                 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2063                                                     prefer,
2064                                                     fl, type, family, dir,
2065                                                     if_id);
2066                 if (!tmp)
2067                         continue;
2068
2069                 if (IS_ERR(tmp))
2070                         return tmp;
2071                 prefer = tmp;
2072         }
2073
2074         return prefer;
2075 }
2076
2077 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2078                                                      const struct flowi *fl,
2079                                                      u16 family, u8 dir,
2080                                                      u32 if_id)
2081 {
2082         struct xfrm_pol_inexact_candidates cand;
2083         const xfrm_address_t *daddr, *saddr;
2084         struct xfrm_pol_inexact_bin *bin;
2085         struct xfrm_policy *pol, *ret;
2086         struct hlist_head *chain;
2087         unsigned int sequence;
2088         int err;
2089
2090         daddr = xfrm_flowi_daddr(fl, family);
2091         saddr = xfrm_flowi_saddr(fl, family);
2092         if (unlikely(!daddr || !saddr))
2093                 return NULL;
2094
2095         rcu_read_lock();
2096  retry:
2097         do {
2098                 sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2099                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2100         } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence));
2101
2102         ret = NULL;
2103         hlist_for_each_entry_rcu(pol, chain, bydst) {
2104                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2105                 if (err) {
2106                         if (err == -ESRCH)
2107                                 continue;
2108                         else {
2109                                 ret = ERR_PTR(err);
2110                                 goto fail;
2111                         }
2112                 } else {
2113                         ret = pol;
2114                         break;
2115                 }
2116         }
2117         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2118         if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2119                                                          daddr))
2120                 goto skip_inexact;
2121
2122         pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2123                                           family, dir, if_id);
2124         if (pol) {
2125                 ret = pol;
2126                 if (IS_ERR(pol))
2127                         goto fail;
2128         }
2129
2130 skip_inexact:
2131         if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence))
2132                 goto retry;
2133
2134         if (ret && !xfrm_pol_hold_rcu(ret))
2135                 goto retry;
2136 fail:
2137         rcu_read_unlock();
2138
2139         return ret;
2140 }
2141
2142 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2143                                               const struct flowi *fl,
2144                                               u16 family, u8 dir, u32 if_id)
2145 {
2146 #ifdef CONFIG_XFRM_SUB_POLICY
2147         struct xfrm_policy *pol;
2148
2149         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2150                                         dir, if_id);
2151         if (pol != NULL)
2152                 return pol;
2153 #endif
2154         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2155                                          dir, if_id);
2156 }
2157
2158 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2159                                                  const struct flowi *fl,
2160                                                  u16 family, u32 if_id)
2161 {
2162         struct xfrm_policy *pol;
2163
2164         rcu_read_lock();
2165  again:
2166         pol = rcu_dereference(sk->sk_policy[dir]);
2167         if (pol != NULL) {
2168                 bool match;
2169                 int err = 0;
2170
2171                 if (pol->family != family) {
2172                         pol = NULL;
2173                         goto out;
2174                 }
2175
2176                 match = xfrm_selector_match(&pol->selector, fl, family);
2177                 if (match) {
2178                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2179                             pol->if_id != if_id) {
2180                                 pol = NULL;
2181                                 goto out;
2182                         }
2183                         err = security_xfrm_policy_lookup(pol->security,
2184                                                       fl->flowi_secid,
2185                                                       dir);
2186                         if (!err) {
2187                                 if (!xfrm_pol_hold_rcu(pol))
2188                                         goto again;
2189                         } else if (err == -ESRCH) {
2190                                 pol = NULL;
2191                         } else {
2192                                 pol = ERR_PTR(err);
2193                         }
2194                 } else
2195                         pol = NULL;
2196         }
2197 out:
2198         rcu_read_unlock();
2199         return pol;
2200 }
2201
2202 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2203 {
2204         struct net *net = xp_net(pol);
2205
2206         list_add(&pol->walk.all, &net->xfrm.policy_all);
2207         net->xfrm.policy_count[dir]++;
2208         xfrm_pol_hold(pol);
2209 }
2210
2211 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2212                                                 int dir)
2213 {
2214         struct net *net = xp_net(pol);
2215
2216         if (list_empty(&pol->walk.all))
2217                 return NULL;
2218
2219         /* Socket policies are not hashed. */
2220         if (!hlist_unhashed(&pol->bydst)) {
2221                 hlist_del_rcu(&pol->bydst);
2222                 hlist_del_init(&pol->bydst_inexact_list);
2223                 hlist_del(&pol->byidx);
2224         }
2225
2226         list_del_init(&pol->walk.all);
2227         net->xfrm.policy_count[dir]--;
2228
2229         return pol;
2230 }
2231
2232 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2233 {
2234         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2235 }
2236
2237 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2238 {
2239         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2240 }
2241
2242 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2243 {
2244         struct net *net = xp_net(pol);
2245
2246         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2247         pol = __xfrm_policy_unlink(pol, dir);
2248         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2249         if (pol) {
2250                 xfrm_policy_kill(pol);
2251                 return 0;
2252         }
2253         return -ENOENT;
2254 }
2255 EXPORT_SYMBOL(xfrm_policy_delete);
2256
2257 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2258 {
2259         struct net *net = sock_net(sk);
2260         struct xfrm_policy *old_pol;
2261
2262 #ifdef CONFIG_XFRM_SUB_POLICY
2263         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2264                 return -EINVAL;
2265 #endif
2266
2267         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2268         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2269                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2270         if (pol) {
2271                 pol->curlft.add_time = ktime_get_real_seconds();
2272                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2273                 xfrm_sk_policy_link(pol, dir);
2274         }
2275         rcu_assign_pointer(sk->sk_policy[dir], pol);
2276         if (old_pol) {
2277                 if (pol)
2278                         xfrm_policy_requeue(old_pol, pol);
2279
2280                 /* Unlinking succeeds always. This is the only function
2281                  * allowed to delete or replace socket policy.
2282                  */
2283                 xfrm_sk_policy_unlink(old_pol, dir);
2284         }
2285         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2286
2287         if (old_pol) {
2288                 xfrm_policy_kill(old_pol);
2289         }
2290         return 0;
2291 }
2292
2293 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2294 {
2295         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2296         struct net *net = xp_net(old);
2297
2298         if (newp) {
2299                 newp->selector = old->selector;
2300                 if (security_xfrm_policy_clone(old->security,
2301                                                &newp->security)) {
2302                         kfree(newp);
2303                         return NULL;  /* ENOMEM */
2304                 }
2305                 newp->lft = old->lft;
2306                 newp->curlft = old->curlft;
2307                 newp->mark = old->mark;
2308                 newp->if_id = old->if_id;
2309                 newp->action = old->action;
2310                 newp->flags = old->flags;
2311                 newp->xfrm_nr = old->xfrm_nr;
2312                 newp->index = old->index;
2313                 newp->type = old->type;
2314                 newp->family = old->family;
2315                 memcpy(newp->xfrm_vec, old->xfrm_vec,
2316                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2317                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2318                 xfrm_sk_policy_link(newp, dir);
2319                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2320                 xfrm_pol_put(newp);
2321         }
2322         return newp;
2323 }
2324
2325 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2326 {
2327         const struct xfrm_policy *p;
2328         struct xfrm_policy *np;
2329         int i, ret = 0;
2330
2331         rcu_read_lock();
2332         for (i = 0; i < 2; i++) {
2333                 p = rcu_dereference(osk->sk_policy[i]);
2334                 if (p) {
2335                         np = clone_policy(p, i);
2336                         if (unlikely(!np)) {
2337                                 ret = -ENOMEM;
2338                                 break;
2339                         }
2340                         rcu_assign_pointer(sk->sk_policy[i], np);
2341                 }
2342         }
2343         rcu_read_unlock();
2344         return ret;
2345 }
2346
2347 static int
2348 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2349                xfrm_address_t *remote, unsigned short family, u32 mark)
2350 {
2351         int err;
2352         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2353
2354         if (unlikely(afinfo == NULL))
2355                 return -EINVAL;
2356         err = afinfo->get_saddr(net, oif, local, remote, mark);
2357         rcu_read_unlock();
2358         return err;
2359 }
2360
2361 /* Resolve list of templates for the flow, given policy. */
2362
2363 static int
2364 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2365                       struct xfrm_state **xfrm, unsigned short family)
2366 {
2367         struct net *net = xp_net(policy);
2368         int nx;
2369         int i, error;
2370         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2371         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2372         xfrm_address_t tmp;
2373
2374         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2375                 struct xfrm_state *x;
2376                 xfrm_address_t *remote = daddr;
2377                 xfrm_address_t *local  = saddr;
2378                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2379
2380                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2381                     tmpl->mode == XFRM_MODE_BEET) {
2382                         remote = &tmpl->id.daddr;
2383                         local = &tmpl->saddr;
2384                         if (xfrm_addr_any(local, tmpl->encap_family)) {
2385                                 error = xfrm_get_saddr(net, fl->flowi_oif,
2386                                                        &tmp, remote,
2387                                                        tmpl->encap_family, 0);
2388                                 if (error)
2389                                         goto fail;
2390                                 local = &tmp;
2391                         }
2392                 }
2393
2394                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2395                                     family, policy->if_id);
2396
2397                 if (x && x->km.state == XFRM_STATE_VALID) {
2398                         xfrm[nx++] = x;
2399                         daddr = remote;
2400                         saddr = local;
2401                         continue;
2402                 }
2403                 if (x) {
2404                         error = (x->km.state == XFRM_STATE_ERROR ?
2405                                  -EINVAL : -EAGAIN);
2406                         xfrm_state_put(x);
2407                 } else if (error == -ESRCH) {
2408                         error = -EAGAIN;
2409                 }
2410
2411                 if (!tmpl->optional)
2412                         goto fail;
2413         }
2414         return nx;
2415
2416 fail:
2417         for (nx--; nx >= 0; nx--)
2418                 xfrm_state_put(xfrm[nx]);
2419         return error;
2420 }
2421
2422 static int
2423 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2424                   struct xfrm_state **xfrm, unsigned short family)
2425 {
2426         struct xfrm_state *tp[XFRM_MAX_DEPTH];
2427         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2428         int cnx = 0;
2429         int error;
2430         int ret;
2431         int i;
2432
2433         for (i = 0; i < npols; i++) {
2434                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2435                         error = -ENOBUFS;
2436                         goto fail;
2437                 }
2438
2439                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2440                 if (ret < 0) {
2441                         error = ret;
2442                         goto fail;
2443                 } else
2444                         cnx += ret;
2445         }
2446
2447         /* found states are sorted for outbound processing */
2448         if (npols > 1)
2449                 xfrm_state_sort(xfrm, tpp, cnx, family);
2450
2451         return cnx;
2452
2453  fail:
2454         for (cnx--; cnx >= 0; cnx--)
2455                 xfrm_state_put(tpp[cnx]);
2456         return error;
2457
2458 }
2459
2460 static int xfrm_get_tos(const struct flowi *fl, int family)
2461 {
2462         if (family == AF_INET)
2463                 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2464
2465         return 0;
2466 }
2467
2468 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2469 {
2470         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2471         struct dst_ops *dst_ops;
2472         struct xfrm_dst *xdst;
2473
2474         if (!afinfo)
2475                 return ERR_PTR(-EINVAL);
2476
2477         switch (family) {
2478         case AF_INET:
2479                 dst_ops = &net->xfrm.xfrm4_dst_ops;
2480                 break;
2481 #if IS_ENABLED(CONFIG_IPV6)
2482         case AF_INET6:
2483                 dst_ops = &net->xfrm.xfrm6_dst_ops;
2484                 break;
2485 #endif
2486         default:
2487                 BUG();
2488         }
2489         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2490
2491         if (likely(xdst)) {
2492                 struct dst_entry *dst = &xdst->u.dst;
2493
2494                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2495         } else
2496                 xdst = ERR_PTR(-ENOBUFS);
2497
2498         rcu_read_unlock();
2499
2500         return xdst;
2501 }
2502
2503 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2504                            int nfheader_len)
2505 {
2506         if (dst->ops->family == AF_INET6) {
2507                 struct rt6_info *rt = (struct rt6_info *)dst;
2508                 path->path_cookie = rt6_get_cookie(rt);
2509                 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2510         }
2511 }
2512
2513 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2514                                 const struct flowi *fl)
2515 {
2516         const struct xfrm_policy_afinfo *afinfo =
2517                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2518         int err;
2519
2520         if (!afinfo)
2521                 return -EINVAL;
2522
2523         err = afinfo->fill_dst(xdst, dev, fl);
2524
2525         rcu_read_unlock();
2526
2527         return err;
2528 }
2529
2530
2531 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2532  * all the metrics... Shortly, bundle a bundle.
2533  */
2534
2535 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2536                                             struct xfrm_state **xfrm,
2537                                             struct xfrm_dst **bundle,
2538                                             int nx,
2539                                             const struct flowi *fl,
2540                                             struct dst_entry *dst)
2541 {
2542         const struct xfrm_state_afinfo *afinfo;
2543         const struct xfrm_mode *inner_mode;
2544         struct net *net = xp_net(policy);
2545         unsigned long now = jiffies;
2546         struct net_device *dev;
2547         struct xfrm_dst *xdst_prev = NULL;
2548         struct xfrm_dst *xdst0 = NULL;
2549         int i = 0;
2550         int err;
2551         int header_len = 0;
2552         int nfheader_len = 0;
2553         int trailer_len = 0;
2554         int tos;
2555         int family = policy->selector.family;
2556         xfrm_address_t saddr, daddr;
2557
2558         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2559
2560         tos = xfrm_get_tos(fl, family);
2561
2562         dst_hold(dst);
2563
2564         for (; i < nx; i++) {
2565                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2566                 struct dst_entry *dst1 = &xdst->u.dst;
2567
2568                 err = PTR_ERR(xdst);
2569                 if (IS_ERR(xdst)) {
2570                         dst_release(dst);
2571                         goto put_states;
2572                 }
2573
2574                 bundle[i] = xdst;
2575                 if (!xdst_prev)
2576                         xdst0 = xdst;
2577                 else
2578                         /* Ref count is taken during xfrm_alloc_dst()
2579                          * No need to do dst_clone() on dst1
2580                          */
2581                         xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2582
2583                 if (xfrm[i]->sel.family == AF_UNSPEC) {
2584                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
2585                                                         xfrm_af2proto(family));
2586                         if (!inner_mode) {
2587                                 err = -EAFNOSUPPORT;
2588                                 dst_release(dst);
2589                                 goto put_states;
2590                         }
2591                 } else
2592                         inner_mode = &xfrm[i]->inner_mode;
2593
2594                 xdst->route = dst;
2595                 dst_copy_metrics(dst1, dst);
2596
2597                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2598                         __u32 mark = 0;
2599
2600                         if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2601                                 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2602
2603                         family = xfrm[i]->props.family;
2604                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2605                                               &saddr, &daddr, family, mark);
2606                         err = PTR_ERR(dst);
2607                         if (IS_ERR(dst))
2608                                 goto put_states;
2609                 } else
2610                         dst_hold(dst);
2611
2612                 dst1->xfrm = xfrm[i];
2613                 xdst->xfrm_genid = xfrm[i]->genid;
2614
2615                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2616                 dst1->lastuse = now;
2617
2618                 dst1->input = dst_discard;
2619
2620                 rcu_read_lock();
2621                 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2622                 if (likely(afinfo))
2623                         dst1->output = afinfo->output;
2624                 else
2625                         dst1->output = dst_discard_out;
2626                 rcu_read_unlock();
2627
2628                 xdst_prev = xdst;
2629
2630                 header_len += xfrm[i]->props.header_len;
2631                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2632                         nfheader_len += xfrm[i]->props.header_len;
2633                 trailer_len += xfrm[i]->props.trailer_len;
2634         }
2635
2636         xfrm_dst_set_child(xdst_prev, dst);
2637         xdst0->path = dst;
2638
2639         err = -ENODEV;
2640         dev = dst->dev;
2641         if (!dev)
2642                 goto free_dst;
2643
2644         xfrm_init_path(xdst0, dst, nfheader_len);
2645         xfrm_init_pmtu(bundle, nx);
2646
2647         for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2648              xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2649                 err = xfrm_fill_dst(xdst_prev, dev, fl);
2650                 if (err)
2651                         goto free_dst;
2652
2653                 xdst_prev->u.dst.header_len = header_len;
2654                 xdst_prev->u.dst.trailer_len = trailer_len;
2655                 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2656                 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2657         }
2658
2659         return &xdst0->u.dst;
2660
2661 put_states:
2662         for (; i < nx; i++)
2663                 xfrm_state_put(xfrm[i]);
2664 free_dst:
2665         if (xdst0)
2666                 dst_release_immediate(&xdst0->u.dst);
2667
2668         return ERR_PTR(err);
2669 }
2670
2671 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2672                                 struct xfrm_policy **pols,
2673                                 int *num_pols, int *num_xfrms)
2674 {
2675         int i;
2676
2677         if (*num_pols == 0 || !pols[0]) {
2678                 *num_pols = 0;
2679                 *num_xfrms = 0;
2680                 return 0;
2681         }
2682         if (IS_ERR(pols[0]))
2683                 return PTR_ERR(pols[0]);
2684
2685         *num_xfrms = pols[0]->xfrm_nr;
2686
2687 #ifdef CONFIG_XFRM_SUB_POLICY
2688         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2689             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2690                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2691                                                     XFRM_POLICY_TYPE_MAIN,
2692                                                     fl, family,
2693                                                     XFRM_POLICY_OUT,
2694                                                     pols[0]->if_id);
2695                 if (pols[1]) {
2696                         if (IS_ERR(pols[1])) {
2697                                 xfrm_pols_put(pols, *num_pols);
2698                                 return PTR_ERR(pols[1]);
2699                         }
2700                         (*num_pols)++;
2701                         (*num_xfrms) += pols[1]->xfrm_nr;
2702                 }
2703         }
2704 #endif
2705         for (i = 0; i < *num_pols; i++) {
2706                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2707                         *num_xfrms = -1;
2708                         break;
2709                 }
2710         }
2711
2712         return 0;
2713
2714 }
2715
2716 static struct xfrm_dst *
2717 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2718                                const struct flowi *fl, u16 family,
2719                                struct dst_entry *dst_orig)
2720 {
2721         struct net *net = xp_net(pols[0]);
2722         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2723         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2724         struct xfrm_dst *xdst;
2725         struct dst_entry *dst;
2726         int err;
2727
2728         /* Try to instantiate a bundle */
2729         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2730         if (err <= 0) {
2731                 if (err == 0)
2732                         return NULL;
2733
2734                 if (err != -EAGAIN)
2735                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2736                 return ERR_PTR(err);
2737         }
2738
2739         dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2740         if (IS_ERR(dst)) {
2741                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2742                 return ERR_CAST(dst);
2743         }
2744
2745         xdst = (struct xfrm_dst *)dst;
2746         xdst->num_xfrms = err;
2747         xdst->num_pols = num_pols;
2748         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2749         xdst->policy_genid = atomic_read(&pols[0]->genid);
2750
2751         return xdst;
2752 }
2753
2754 static void xfrm_policy_queue_process(struct timer_list *t)
2755 {
2756         struct sk_buff *skb;
2757         struct sock *sk;
2758         struct dst_entry *dst;
2759         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2760         struct net *net = xp_net(pol);
2761         struct xfrm_policy_queue *pq = &pol->polq;
2762         struct flowi fl;
2763         struct sk_buff_head list;
2764         __u32 skb_mark;
2765
2766         spin_lock(&pq->hold_queue.lock);
2767         skb = skb_peek(&pq->hold_queue);
2768         if (!skb) {
2769                 spin_unlock(&pq->hold_queue.lock);
2770                 goto out;
2771         }
2772         dst = skb_dst(skb);
2773         sk = skb->sk;
2774
2775         /* Fixup the mark to support VTI. */
2776         skb_mark = skb->mark;
2777         skb->mark = pol->mark.v;
2778         xfrm_decode_session(skb, &fl, dst->ops->family);
2779         skb->mark = skb_mark;
2780         spin_unlock(&pq->hold_queue.lock);
2781
2782         dst_hold(xfrm_dst_path(dst));
2783         dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2784         if (IS_ERR(dst))
2785                 goto purge_queue;
2786
2787         if (dst->flags & DST_XFRM_QUEUE) {
2788                 dst_release(dst);
2789
2790                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2791                         goto purge_queue;
2792
2793                 pq->timeout = pq->timeout << 1;
2794                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2795                         xfrm_pol_hold(pol);
2796                 goto out;
2797         }
2798
2799         dst_release(dst);
2800
2801         __skb_queue_head_init(&list);
2802
2803         spin_lock(&pq->hold_queue.lock);
2804         pq->timeout = 0;
2805         skb_queue_splice_init(&pq->hold_queue, &list);
2806         spin_unlock(&pq->hold_queue.lock);
2807
2808         while (!skb_queue_empty(&list)) {
2809                 skb = __skb_dequeue(&list);
2810
2811                 /* Fixup the mark to support VTI. */
2812                 skb_mark = skb->mark;
2813                 skb->mark = pol->mark.v;
2814                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2815                 skb->mark = skb_mark;
2816
2817                 dst_hold(xfrm_dst_path(skb_dst(skb)));
2818                 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2819                 if (IS_ERR(dst)) {
2820                         kfree_skb(skb);
2821                         continue;
2822                 }
2823
2824                 nf_reset_ct(skb);
2825                 skb_dst_drop(skb);
2826                 skb_dst_set(skb, dst);
2827
2828                 dst_output(net, skb->sk, skb);
2829         }
2830
2831 out:
2832         xfrm_pol_put(pol);
2833         return;
2834
2835 purge_queue:
2836         pq->timeout = 0;
2837         skb_queue_purge(&pq->hold_queue);
2838         xfrm_pol_put(pol);
2839 }
2840
2841 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2842 {
2843         unsigned long sched_next;
2844         struct dst_entry *dst = skb_dst(skb);
2845         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2846         struct xfrm_policy *pol = xdst->pols[0];
2847         struct xfrm_policy_queue *pq = &pol->polq;
2848
2849         if (unlikely(skb_fclone_busy(sk, skb))) {
2850                 kfree_skb(skb);
2851                 return 0;
2852         }
2853
2854         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2855                 kfree_skb(skb);
2856                 return -EAGAIN;
2857         }
2858
2859         skb_dst_force(skb);
2860
2861         spin_lock_bh(&pq->hold_queue.lock);
2862
2863         if (!pq->timeout)
2864                 pq->timeout = XFRM_QUEUE_TMO_MIN;
2865
2866         sched_next = jiffies + pq->timeout;
2867
2868         if (del_timer(&pq->hold_timer)) {
2869                 if (time_before(pq->hold_timer.expires, sched_next))
2870                         sched_next = pq->hold_timer.expires;
2871                 xfrm_pol_put(pol);
2872         }
2873
2874         __skb_queue_tail(&pq->hold_queue, skb);
2875         if (!mod_timer(&pq->hold_timer, sched_next))
2876                 xfrm_pol_hold(pol);
2877
2878         spin_unlock_bh(&pq->hold_queue.lock);
2879
2880         return 0;
2881 }
2882
2883 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2884                                                  struct xfrm_flo *xflo,
2885                                                  const struct flowi *fl,
2886                                                  int num_xfrms,
2887                                                  u16 family)
2888 {
2889         int err;
2890         struct net_device *dev;
2891         struct dst_entry *dst;
2892         struct dst_entry *dst1;
2893         struct xfrm_dst *xdst;
2894
2895         xdst = xfrm_alloc_dst(net, family);
2896         if (IS_ERR(xdst))
2897                 return xdst;
2898
2899         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2900             net->xfrm.sysctl_larval_drop ||
2901             num_xfrms <= 0)
2902                 return xdst;
2903
2904         dst = xflo->dst_orig;
2905         dst1 = &xdst->u.dst;
2906         dst_hold(dst);
2907         xdst->route = dst;
2908
2909         dst_copy_metrics(dst1, dst);
2910
2911         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2912         dst1->flags |= DST_XFRM_QUEUE;
2913         dst1->lastuse = jiffies;
2914
2915         dst1->input = dst_discard;
2916         dst1->output = xdst_queue_output;
2917
2918         dst_hold(dst);
2919         xfrm_dst_set_child(xdst, dst);
2920         xdst->path = dst;
2921
2922         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2923
2924         err = -ENODEV;
2925         dev = dst->dev;
2926         if (!dev)
2927                 goto free_dst;
2928
2929         err = xfrm_fill_dst(xdst, dev, fl);
2930         if (err)
2931                 goto free_dst;
2932
2933 out:
2934         return xdst;
2935
2936 free_dst:
2937         dst_release(dst1);
2938         xdst = ERR_PTR(err);
2939         goto out;
2940 }
2941
2942 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2943                                            const struct flowi *fl,
2944                                            u16 family, u8 dir,
2945                                            struct xfrm_flo *xflo, u32 if_id)
2946 {
2947         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2948         int num_pols = 0, num_xfrms = 0, err;
2949         struct xfrm_dst *xdst;
2950
2951         /* Resolve policies to use if we couldn't get them from
2952          * previous cache entry */
2953         num_pols = 1;
2954         pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2955         err = xfrm_expand_policies(fl, family, pols,
2956                                            &num_pols, &num_xfrms);
2957         if (err < 0)
2958                 goto inc_error;
2959         if (num_pols == 0)
2960                 return NULL;
2961         if (num_xfrms <= 0)
2962                 goto make_dummy_bundle;
2963
2964         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2965                                               xflo->dst_orig);
2966         if (IS_ERR(xdst)) {
2967                 err = PTR_ERR(xdst);
2968                 if (err == -EREMOTE) {
2969                         xfrm_pols_put(pols, num_pols);
2970                         return NULL;
2971                 }
2972
2973                 if (err != -EAGAIN)
2974                         goto error;
2975                 goto make_dummy_bundle;
2976         } else if (xdst == NULL) {
2977                 num_xfrms = 0;
2978                 goto make_dummy_bundle;
2979         }
2980
2981         return xdst;
2982
2983 make_dummy_bundle:
2984         /* We found policies, but there's no bundles to instantiate:
2985          * either because the policy blocks, has no transformations or
2986          * we could not build template (no xfrm_states).*/
2987         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2988         if (IS_ERR(xdst)) {
2989                 xfrm_pols_put(pols, num_pols);
2990                 return ERR_CAST(xdst);
2991         }
2992         xdst->num_pols = num_pols;
2993         xdst->num_xfrms = num_xfrms;
2994         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2995
2996         return xdst;
2997
2998 inc_error:
2999         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3000 error:
3001         xfrm_pols_put(pols, num_pols);
3002         return ERR_PTR(err);
3003 }
3004
3005 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3006                                         struct dst_entry *dst_orig)
3007 {
3008         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3009         struct dst_entry *ret;
3010
3011         if (!afinfo) {
3012                 dst_release(dst_orig);
3013                 return ERR_PTR(-EINVAL);
3014         } else {
3015                 ret = afinfo->blackhole_route(net, dst_orig);
3016         }
3017         rcu_read_unlock();
3018
3019         return ret;
3020 }
3021
3022 /* Finds/creates a bundle for given flow and if_id
3023  *
3024  * At the moment we eat a raw IP route. Mostly to speed up lookups
3025  * on interfaces with disabled IPsec.
3026  *
3027  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3028  * compatibility
3029  */
3030 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3031                                         struct dst_entry *dst_orig,
3032                                         const struct flowi *fl,
3033                                         const struct sock *sk,
3034                                         int flags, u32 if_id)
3035 {
3036         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3037         struct xfrm_dst *xdst;
3038         struct dst_entry *dst, *route;
3039         u16 family = dst_orig->ops->family;
3040         u8 dir = XFRM_POLICY_OUT;
3041         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3042
3043         dst = NULL;
3044         xdst = NULL;
3045         route = NULL;
3046
3047         sk = sk_const_to_full_sk(sk);
3048         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3049                 num_pols = 1;
3050                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3051                                                 if_id);
3052                 err = xfrm_expand_policies(fl, family, pols,
3053                                            &num_pols, &num_xfrms);
3054                 if (err < 0)
3055                         goto dropdst;
3056
3057                 if (num_pols) {
3058                         if (num_xfrms <= 0) {
3059                                 drop_pols = num_pols;
3060                                 goto no_transform;
3061                         }
3062
3063                         xdst = xfrm_resolve_and_create_bundle(
3064                                         pols, num_pols, fl,
3065                                         family, dst_orig);
3066
3067                         if (IS_ERR(xdst)) {
3068                                 xfrm_pols_put(pols, num_pols);
3069                                 err = PTR_ERR(xdst);
3070                                 if (err == -EREMOTE)
3071                                         goto nopol;
3072
3073                                 goto dropdst;
3074                         } else if (xdst == NULL) {
3075                                 num_xfrms = 0;
3076                                 drop_pols = num_pols;
3077                                 goto no_transform;
3078                         }
3079
3080                         route = xdst->route;
3081                 }
3082         }
3083
3084         if (xdst == NULL) {
3085                 struct xfrm_flo xflo;
3086
3087                 xflo.dst_orig = dst_orig;
3088                 xflo.flags = flags;
3089
3090                 /* To accelerate a bit...  */
3091                 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3092                                !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3093                         goto nopol;
3094
3095                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3096                 if (xdst == NULL)
3097                         goto nopol;
3098                 if (IS_ERR(xdst)) {
3099                         err = PTR_ERR(xdst);
3100                         goto dropdst;
3101                 }
3102
3103                 num_pols = xdst->num_pols;
3104                 num_xfrms = xdst->num_xfrms;
3105                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3106                 route = xdst->route;
3107         }
3108
3109         dst = &xdst->u.dst;
3110         if (route == NULL && num_xfrms > 0) {
3111                 /* The only case when xfrm_bundle_lookup() returns a
3112                  * bundle with null route, is when the template could
3113                  * not be resolved. It means policies are there, but
3114                  * bundle could not be created, since we don't yet
3115                  * have the xfrm_state's. We need to wait for KM to
3116                  * negotiate new SA's or bail out with error.*/
3117                 if (net->xfrm.sysctl_larval_drop) {
3118                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3119                         err = -EREMOTE;
3120                         goto error;
3121                 }
3122
3123                 err = -EAGAIN;
3124
3125                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3126                 goto error;
3127         }
3128
3129 no_transform:
3130         if (num_pols == 0)
3131                 goto nopol;
3132
3133         if ((flags & XFRM_LOOKUP_ICMP) &&
3134             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3135                 err = -ENOENT;
3136                 goto error;
3137         }
3138
3139         for (i = 0; i < num_pols; i++)
3140                 pols[i]->curlft.use_time = ktime_get_real_seconds();
3141
3142         if (num_xfrms < 0) {
3143                 /* Prohibit the flow */
3144                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3145                 err = -EPERM;
3146                 goto error;
3147         } else if (num_xfrms > 0) {
3148                 /* Flow transformed */
3149                 dst_release(dst_orig);
3150         } else {
3151                 /* Flow passes untransformed */
3152                 dst_release(dst);
3153                 dst = dst_orig;
3154         }
3155 ok:
3156         xfrm_pols_put(pols, drop_pols);
3157         if (dst && dst->xfrm &&
3158             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3159                 dst->flags |= DST_XFRM_TUNNEL;
3160         return dst;
3161
3162 nopol:
3163         if (!(flags & XFRM_LOOKUP_ICMP)) {
3164                 dst = dst_orig;
3165                 goto ok;
3166         }
3167         err = -ENOENT;
3168 error:
3169         dst_release(dst);
3170 dropdst:
3171         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3172                 dst_release(dst_orig);
3173         xfrm_pols_put(pols, drop_pols);
3174         return ERR_PTR(err);
3175 }
3176 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3177
3178 /* Main function: finds/creates a bundle for given flow.
3179  *
3180  * At the moment we eat a raw IP route. Mostly to speed up lookups
3181  * on interfaces with disabled IPsec.
3182  */
3183 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3184                               const struct flowi *fl, const struct sock *sk,
3185                               int flags)
3186 {
3187         return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3188 }
3189 EXPORT_SYMBOL(xfrm_lookup);
3190
3191 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3192  * Otherwise we may send out blackholed packets.
3193  */
3194 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3195                                     const struct flowi *fl,
3196                                     const struct sock *sk, int flags)
3197 {
3198         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3199                                             flags | XFRM_LOOKUP_QUEUE |
3200                                             XFRM_LOOKUP_KEEP_DST_REF);
3201
3202         if (PTR_ERR(dst) == -EREMOTE)
3203                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3204
3205         if (IS_ERR(dst))
3206                 dst_release(dst_orig);
3207
3208         return dst;
3209 }
3210 EXPORT_SYMBOL(xfrm_lookup_route);
3211
3212 static inline int
3213 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3214 {
3215         struct sec_path *sp = skb_sec_path(skb);
3216         struct xfrm_state *x;
3217
3218         if (!sp || idx < 0 || idx >= sp->len)
3219                 return 0;
3220         x = sp->xvec[idx];
3221         if (!x->type->reject)
3222                 return 0;
3223         return x->type->reject(x, skb, fl);
3224 }
3225
3226 /* When skb is transformed back to its "native" form, we have to
3227  * check policy restrictions. At the moment we make this in maximally
3228  * stupid way. Shame on me. :-) Of course, connected sockets must
3229  * have policy cached at them.
3230  */
3231
3232 static inline int
3233 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3234               unsigned short family)
3235 {
3236         if (xfrm_state_kern(x))
3237                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3238         return  x->id.proto == tmpl->id.proto &&
3239                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3240                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3241                 x->props.mode == tmpl->mode &&
3242                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3243                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3244                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3245                   xfrm_state_addr_cmp(tmpl, x, family));
3246 }
3247
3248 /*
3249  * 0 or more than 0 is returned when validation is succeeded (either bypass
3250  * because of optional transport mode, or next index of the mathced secpath
3251  * state with the template.
3252  * -1 is returned when no matching template is found.
3253  * Otherwise "-2 - errored_index" is returned.
3254  */
3255 static inline int
3256 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3257                unsigned short family)
3258 {
3259         int idx = start;
3260
3261         if (tmpl->optional) {
3262                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3263                         return start;
3264         } else
3265                 start = -1;
3266         for (; idx < sp->len; idx++) {
3267                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3268                         return ++idx;
3269                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3270                         if (start == -1)
3271                                 start = -2-idx;
3272                         break;
3273                 }
3274         }
3275         return start;
3276 }
3277
3278 static void
3279 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3280 {
3281         const struct iphdr *iph = ip_hdr(skb);
3282         int ihl = iph->ihl;
3283         u8 *xprth = skb_network_header(skb) + ihl * 4;
3284         struct flowi4 *fl4 = &fl->u.ip4;
3285         int oif = 0;
3286
3287         if (skb_dst(skb) && skb_dst(skb)->dev)
3288                 oif = skb_dst(skb)->dev->ifindex;
3289
3290         memset(fl4, 0, sizeof(struct flowi4));
3291         fl4->flowi4_mark = skb->mark;
3292         fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3293
3294         fl4->flowi4_proto = iph->protocol;
3295         fl4->daddr = reverse ? iph->saddr : iph->daddr;
3296         fl4->saddr = reverse ? iph->daddr : iph->saddr;
3297         fl4->flowi4_tos = iph->tos;
3298
3299         if (!ip_is_fragment(iph)) {
3300                 switch (iph->protocol) {
3301                 case IPPROTO_UDP:
3302                 case IPPROTO_UDPLITE:
3303                 case IPPROTO_TCP:
3304                 case IPPROTO_SCTP:
3305                 case IPPROTO_DCCP:
3306                         if (xprth + 4 < skb->data ||
3307                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
3308                                 __be16 *ports;
3309
3310                                 xprth = skb_network_header(skb) + ihl * 4;
3311                                 ports = (__be16 *)xprth;
3312
3313                                 fl4->fl4_sport = ports[!!reverse];
3314                                 fl4->fl4_dport = ports[!reverse];
3315                         }
3316                         break;
3317                 case IPPROTO_ICMP:
3318                         if (xprth + 2 < skb->data ||
3319                             pskb_may_pull(skb, xprth + 2 - skb->data)) {
3320                                 u8 *icmp;
3321
3322                                 xprth = skb_network_header(skb) + ihl * 4;
3323                                 icmp = xprth;
3324
3325                                 fl4->fl4_icmp_type = icmp[0];
3326                                 fl4->fl4_icmp_code = icmp[1];
3327                         }
3328                         break;
3329                 case IPPROTO_GRE:
3330                         if (xprth + 12 < skb->data ||
3331                             pskb_may_pull(skb, xprth + 12 - skb->data)) {
3332                                 __be16 *greflags;
3333                                 __be32 *gre_hdr;
3334
3335                                 xprth = skb_network_header(skb) + ihl * 4;
3336                                 greflags = (__be16 *)xprth;
3337                                 gre_hdr = (__be32 *)xprth;
3338
3339                                 if (greflags[0] & GRE_KEY) {
3340                                         if (greflags[0] & GRE_CSUM)
3341                                                 gre_hdr++;
3342                                         fl4->fl4_gre_key = gre_hdr[1];
3343                                 }
3344                         }
3345                         break;
3346                 default:
3347                         break;
3348                 }
3349         }
3350 }
3351
3352 #if IS_ENABLED(CONFIG_IPV6)
3353 static void
3354 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3355 {
3356         struct flowi6 *fl6 = &fl->u.ip6;
3357         int onlyproto = 0;
3358         const struct ipv6hdr *hdr = ipv6_hdr(skb);
3359         u32 offset = sizeof(*hdr);
3360         struct ipv6_opt_hdr *exthdr;
3361         const unsigned char *nh = skb_network_header(skb);
3362         u16 nhoff = IP6CB(skb)->nhoff;
3363         int oif = 0;
3364         u8 nexthdr;
3365
3366         if (!nhoff)
3367                 nhoff = offsetof(struct ipv6hdr, nexthdr);
3368
3369         nexthdr = nh[nhoff];
3370
3371         if (skb_dst(skb) && skb_dst(skb)->dev)
3372                 oif = skb_dst(skb)->dev->ifindex;
3373
3374         memset(fl6, 0, sizeof(struct flowi6));
3375         fl6->flowi6_mark = skb->mark;
3376         fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3377
3378         fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3379         fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3380
3381         while (nh + offset + sizeof(*exthdr) < skb->data ||
3382                pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3383                 nh = skb_network_header(skb);
3384                 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3385
3386                 switch (nexthdr) {
3387                 case NEXTHDR_FRAGMENT:
3388                         onlyproto = 1;
3389                         fallthrough;
3390                 case NEXTHDR_ROUTING:
3391                 case NEXTHDR_HOP:
3392                 case NEXTHDR_DEST:
3393                         offset += ipv6_optlen(exthdr);
3394                         nexthdr = exthdr->nexthdr;
3395                         exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3396                         break;
3397                 case IPPROTO_UDP:
3398                 case IPPROTO_UDPLITE:
3399                 case IPPROTO_TCP:
3400                 case IPPROTO_SCTP:
3401                 case IPPROTO_DCCP:
3402                         if (!onlyproto && (nh + offset + 4 < skb->data ||
3403                              pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3404                                 __be16 *ports;
3405
3406                                 nh = skb_network_header(skb);
3407                                 ports = (__be16 *)(nh + offset);
3408                                 fl6->fl6_sport = ports[!!reverse];
3409                                 fl6->fl6_dport = ports[!reverse];
3410                         }
3411                         fl6->flowi6_proto = nexthdr;
3412                         return;
3413                 case IPPROTO_ICMPV6:
3414                         if (!onlyproto && (nh + offset + 2 < skb->data ||
3415                             pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3416                                 u8 *icmp;
3417
3418                                 nh = skb_network_header(skb);
3419                                 icmp = (u8 *)(nh + offset);
3420                                 fl6->fl6_icmp_type = icmp[0];
3421                                 fl6->fl6_icmp_code = icmp[1];
3422                         }
3423                         fl6->flowi6_proto = nexthdr;
3424                         return;
3425 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3426                 case IPPROTO_MH:
3427                         offset += ipv6_optlen(exthdr);
3428                         if (!onlyproto && (nh + offset + 3 < skb->data ||
3429                             pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3430                                 struct ip6_mh *mh;
3431
3432                                 nh = skb_network_header(skb);
3433                                 mh = (struct ip6_mh *)(nh + offset);
3434                                 fl6->fl6_mh_type = mh->ip6mh_type;
3435                         }
3436                         fl6->flowi6_proto = nexthdr;
3437                         return;
3438 #endif
3439                 default:
3440                         fl6->flowi6_proto = nexthdr;
3441                         return;
3442                 }
3443         }
3444 }
3445 #endif
3446
3447 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3448                           unsigned int family, int reverse)
3449 {
3450         switch (family) {
3451         case AF_INET:
3452                 decode_session4(skb, fl, reverse);
3453                 break;
3454 #if IS_ENABLED(CONFIG_IPV6)
3455         case AF_INET6:
3456                 decode_session6(skb, fl, reverse);
3457                 break;
3458 #endif
3459         default:
3460                 return -EAFNOSUPPORT;
3461         }
3462
3463         return security_xfrm_decode_session(skb, &fl->flowi_secid);
3464 }
3465 EXPORT_SYMBOL(__xfrm_decode_session);
3466
3467 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3468 {
3469         for (; k < sp->len; k++) {
3470                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3471                         *idxp = k;
3472                         return 1;
3473                 }
3474         }
3475
3476         return 0;
3477 }
3478
3479 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3480                         unsigned short family)
3481 {
3482         struct net *net = dev_net(skb->dev);
3483         struct xfrm_policy *pol;
3484         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3485         int npols = 0;
3486         int xfrm_nr;
3487         int pi;
3488         int reverse;
3489         struct flowi fl;
3490         int xerr_idx = -1;
3491         const struct xfrm_if_cb *ifcb;
3492         struct sec_path *sp;
3493         struct xfrm_if *xi;
3494         u32 if_id = 0;
3495
3496         rcu_read_lock();
3497         ifcb = xfrm_if_get_cb();
3498
3499         if (ifcb) {
3500                 xi = ifcb->decode_session(skb, family);
3501                 if (xi) {
3502                         if_id = xi->p.if_id;
3503                         net = xi->net;
3504                 }
3505         }
3506         rcu_read_unlock();
3507
3508         reverse = dir & ~XFRM_POLICY_MASK;
3509         dir &= XFRM_POLICY_MASK;
3510
3511         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3512                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3513                 return 0;
3514         }
3515
3516         nf_nat_decode_session(skb, &fl, family);
3517
3518         /* First, check used SA against their selectors. */
3519         sp = skb_sec_path(skb);
3520         if (sp) {
3521                 int i;
3522
3523                 for (i = sp->len - 1; i >= 0; i--) {
3524                         struct xfrm_state *x = sp->xvec[i];
3525                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
3526                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3527                                 return 0;
3528                         }
3529                 }
3530         }
3531
3532         pol = NULL;
3533         sk = sk_to_full_sk(sk);
3534         if (sk && sk->sk_policy[dir]) {
3535                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3536                 if (IS_ERR(pol)) {
3537                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3538                         return 0;
3539                 }
3540         }
3541
3542         if (!pol)
3543                 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3544
3545         if (IS_ERR(pol)) {
3546                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3547                 return 0;
3548         }
3549
3550         if (!pol) {
3551                 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3552                         xfrm_secpath_reject(xerr_idx, skb, &fl);
3553                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3554                         return 0;
3555                 }
3556                 return 1;
3557         }
3558
3559         pol->curlft.use_time = ktime_get_real_seconds();
3560
3561         pols[0] = pol;
3562         npols++;
3563 #ifdef CONFIG_XFRM_SUB_POLICY
3564         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3565                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3566                                                     &fl, family,
3567                                                     XFRM_POLICY_IN, if_id);
3568                 if (pols[1]) {
3569                         if (IS_ERR(pols[1])) {
3570                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3571                                 return 0;
3572                         }
3573                         pols[1]->curlft.use_time = ktime_get_real_seconds();
3574                         npols++;
3575                 }
3576         }
3577 #endif
3578
3579         if (pol->action == XFRM_POLICY_ALLOW) {
3580                 static struct sec_path dummy;
3581                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3582                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3583                 struct xfrm_tmpl **tpp = tp;
3584                 int ti = 0;
3585                 int i, k;
3586
3587                 sp = skb_sec_path(skb);
3588                 if (!sp)
3589                         sp = &dummy;
3590
3591                 for (pi = 0; pi < npols; pi++) {
3592                         if (pols[pi] != pol &&
3593                             pols[pi]->action != XFRM_POLICY_ALLOW) {
3594                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3595                                 goto reject;
3596                         }
3597                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3598                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3599                                 goto reject_error;
3600                         }
3601                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
3602                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3603                 }
3604                 xfrm_nr = ti;
3605                 if (npols > 1) {
3606                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3607                         tpp = stp;
3608                 }
3609
3610                 /* For each tunnel xfrm, find the first matching tmpl.
3611                  * For each tmpl before that, find corresponding xfrm.
3612                  * Order is _important_. Later we will implement
3613                  * some barriers, but at the moment barriers
3614                  * are implied between each two transformations.
3615                  */
3616                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3617                         k = xfrm_policy_ok(tpp[i], sp, k, family);
3618                         if (k < 0) {
3619                                 if (k < -1)
3620                                         /* "-2 - errored_index" returned */
3621                                         xerr_idx = -(2+k);
3622                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3623                                 goto reject;
3624                         }
3625                 }
3626
3627                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3628                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3629                         goto reject;
3630                 }
3631
3632                 xfrm_pols_put(pols, npols);
3633                 return 1;
3634         }
3635         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3636
3637 reject:
3638         xfrm_secpath_reject(xerr_idx, skb, &fl);
3639 reject_error:
3640         xfrm_pols_put(pols, npols);
3641         return 0;
3642 }
3643 EXPORT_SYMBOL(__xfrm_policy_check);
3644
3645 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3646 {
3647         struct net *net = dev_net(skb->dev);
3648         struct flowi fl;
3649         struct dst_entry *dst;
3650         int res = 1;
3651
3652         if (xfrm_decode_session(skb, &fl, family) < 0) {
3653                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3654                 return 0;
3655         }
3656
3657         skb_dst_force(skb);
3658         if (!skb_dst(skb)) {
3659                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3660                 return 0;
3661         }
3662
3663         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3664         if (IS_ERR(dst)) {
3665                 res = 0;
3666                 dst = NULL;
3667         }
3668         skb_dst_set(skb, dst);
3669         return res;
3670 }
3671 EXPORT_SYMBOL(__xfrm_route_forward);
3672
3673 /* Optimize later using cookies and generation ids. */
3674
3675 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3676 {
3677         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3678          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3679          * get validated by dst_ops->check on every use.  We do this
3680          * because when a normal route referenced by an XFRM dst is
3681          * obsoleted we do not go looking around for all parent
3682          * referencing XFRM dsts so that we can invalidate them.  It
3683          * is just too much work.  Instead we make the checks here on
3684          * every use.  For example:
3685          *
3686          *      XFRM dst A --> IPv4 dst X
3687          *
3688          * X is the "xdst->route" of A (X is also the "dst->path" of A
3689          * in this example).  If X is marked obsolete, "A" will not
3690          * notice.  That's what we are validating here via the
3691          * stale_bundle() check.
3692          *
3693          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3694          * be marked on it.
3695          * This will force stale_bundle() to fail on any xdst bundle with
3696          * this dst linked in it.
3697          */
3698         if (dst->obsolete < 0 && !stale_bundle(dst))
3699                 return dst;
3700
3701         return NULL;
3702 }
3703
3704 static int stale_bundle(struct dst_entry *dst)
3705 {
3706         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3707 }
3708
3709 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3710 {
3711         while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3712                 dst->dev = dev_net(dev)->loopback_dev;
3713                 dev_hold(dst->dev);
3714                 dev_put(dev);
3715         }
3716 }
3717 EXPORT_SYMBOL(xfrm_dst_ifdown);
3718
3719 static void xfrm_link_failure(struct sk_buff *skb)
3720 {
3721         /* Impossible. Such dst must be popped before reaches point of failure. */
3722 }
3723
3724 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3725 {
3726         if (dst) {
3727                 if (dst->obsolete) {
3728                         dst_release(dst);
3729                         dst = NULL;
3730                 }
3731         }
3732         return dst;
3733 }
3734
3735 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3736 {
3737         while (nr--) {
3738                 struct xfrm_dst *xdst = bundle[nr];
3739                 u32 pmtu, route_mtu_cached;
3740                 struct dst_entry *dst;
3741
3742                 dst = &xdst->u.dst;
3743                 pmtu = dst_mtu(xfrm_dst_child(dst));
3744                 xdst->child_mtu_cached = pmtu;
3745
3746                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3747
3748                 route_mtu_cached = dst_mtu(xdst->route);
3749                 xdst->route_mtu_cached = route_mtu_cached;
3750
3751                 if (pmtu > route_mtu_cached)
3752                         pmtu = route_mtu_cached;
3753
3754                 dst_metric_set(dst, RTAX_MTU, pmtu);
3755         }
3756 }
3757
3758 /* Check that the bundle accepts the flow and its components are
3759  * still valid.
3760  */
3761
3762 static int xfrm_bundle_ok(struct xfrm_dst *first)
3763 {
3764         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3765         struct dst_entry *dst = &first->u.dst;
3766         struct xfrm_dst *xdst;
3767         int start_from, nr;
3768         u32 mtu;
3769
3770         if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3771             (dst->dev && !netif_running(dst->dev)))
3772                 return 0;
3773
3774         if (dst->flags & DST_XFRM_QUEUE)
3775                 return 1;
3776
3777         start_from = nr = 0;
3778         do {
3779                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3780
3781                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3782                         return 0;
3783                 if (xdst->xfrm_genid != dst->xfrm->genid)
3784                         return 0;
3785                 if (xdst->num_pols > 0 &&
3786                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3787                         return 0;
3788
3789                 bundle[nr++] = xdst;
3790
3791                 mtu = dst_mtu(xfrm_dst_child(dst));
3792                 if (xdst->child_mtu_cached != mtu) {
3793                         start_from = nr;
3794                         xdst->child_mtu_cached = mtu;
3795                 }
3796
3797                 if (!dst_check(xdst->route, xdst->route_cookie))
3798                         return 0;
3799                 mtu = dst_mtu(xdst->route);
3800                 if (xdst->route_mtu_cached != mtu) {
3801                         start_from = nr;
3802                         xdst->route_mtu_cached = mtu;
3803                 }
3804
3805                 dst = xfrm_dst_child(dst);
3806         } while (dst->xfrm);
3807
3808         if (likely(!start_from))
3809                 return 1;
3810
3811         xdst = bundle[start_from - 1];
3812         mtu = xdst->child_mtu_cached;
3813         while (start_from--) {
3814                 dst = &xdst->u.dst;
3815
3816                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3817                 if (mtu > xdst->route_mtu_cached)
3818                         mtu = xdst->route_mtu_cached;
3819                 dst_metric_set(dst, RTAX_MTU, mtu);
3820                 if (!start_from)
3821                         break;
3822
3823                 xdst = bundle[start_from - 1];
3824                 xdst->child_mtu_cached = mtu;
3825         }
3826
3827         return 1;
3828 }
3829
3830 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3831 {
3832         return dst_metric_advmss(xfrm_dst_path(dst));
3833 }
3834
3835 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3836 {
3837         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3838
3839         return mtu ? : dst_mtu(xfrm_dst_path(dst));
3840 }
3841
3842 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3843                                         const void *daddr)
3844 {
3845         while (dst->xfrm) {
3846                 const struct xfrm_state *xfrm = dst->xfrm;
3847
3848                 dst = xfrm_dst_child(dst);
3849
3850                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3851                         continue;
3852                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3853                         daddr = xfrm->coaddr;
3854                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3855                         daddr = &xfrm->id.daddr;
3856         }
3857         return daddr;
3858 }
3859
3860 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3861                                            struct sk_buff *skb,
3862                                            const void *daddr)
3863 {
3864         const struct dst_entry *path = xfrm_dst_path(dst);
3865
3866         if (!skb)
3867                 daddr = xfrm_get_dst_nexthop(dst, daddr);
3868         return path->ops->neigh_lookup(path, skb, daddr);
3869 }
3870
3871 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3872 {
3873         const struct dst_entry *path = xfrm_dst_path(dst);
3874
3875         daddr = xfrm_get_dst_nexthop(dst, daddr);
3876         path->ops->confirm_neigh(path, daddr);
3877 }
3878
3879 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3880 {
3881         int err = 0;
3882
3883         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3884                 return -EAFNOSUPPORT;
3885
3886         spin_lock(&xfrm_policy_afinfo_lock);
3887         if (unlikely(xfrm_policy_afinfo[family] != NULL))
3888                 err = -EEXIST;
3889         else {
3890                 struct dst_ops *dst_ops = afinfo->dst_ops;
3891                 if (likely(dst_ops->kmem_cachep == NULL))
3892                         dst_ops->kmem_cachep = xfrm_dst_cache;
3893                 if (likely(dst_ops->check == NULL))
3894                         dst_ops->check = xfrm_dst_check;
3895                 if (likely(dst_ops->default_advmss == NULL))
3896                         dst_ops->default_advmss = xfrm_default_advmss;
3897                 if (likely(dst_ops->mtu == NULL))
3898                         dst_ops->mtu = xfrm_mtu;
3899                 if (likely(dst_ops->negative_advice == NULL))
3900                         dst_ops->negative_advice = xfrm_negative_advice;
3901                 if (likely(dst_ops->link_failure == NULL))
3902                         dst_ops->link_failure = xfrm_link_failure;
3903                 if (likely(dst_ops->neigh_lookup == NULL))
3904                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
3905                 if (likely(!dst_ops->confirm_neigh))
3906                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
3907                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3908         }
3909         spin_unlock(&xfrm_policy_afinfo_lock);
3910
3911         return err;
3912 }
3913 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3914
3915 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3916 {
3917         struct dst_ops *dst_ops = afinfo->dst_ops;
3918         int i;
3919
3920         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3921                 if (xfrm_policy_afinfo[i] != afinfo)
3922                         continue;
3923                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3924                 break;
3925         }
3926
3927         synchronize_rcu();
3928
3929         dst_ops->kmem_cachep = NULL;
3930         dst_ops->check = NULL;
3931         dst_ops->negative_advice = NULL;
3932         dst_ops->link_failure = NULL;
3933 }
3934 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3935
3936 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3937 {
3938         spin_lock(&xfrm_if_cb_lock);
3939         rcu_assign_pointer(xfrm_if_cb, ifcb);
3940         spin_unlock(&xfrm_if_cb_lock);
3941 }
3942 EXPORT_SYMBOL(xfrm_if_register_cb);
3943
3944 void xfrm_if_unregister_cb(void)
3945 {
3946         RCU_INIT_POINTER(xfrm_if_cb, NULL);
3947         synchronize_rcu();
3948 }
3949 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3950
3951 #ifdef CONFIG_XFRM_STATISTICS
3952 static int __net_init xfrm_statistics_init(struct net *net)
3953 {
3954         int rv;
3955         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3956         if (!net->mib.xfrm_statistics)
3957                 return -ENOMEM;
3958         rv = xfrm_proc_init(net);
3959         if (rv < 0)
3960                 free_percpu(net->mib.xfrm_statistics);
3961         return rv;
3962 }
3963
3964 static void xfrm_statistics_fini(struct net *net)
3965 {
3966         xfrm_proc_fini(net);
3967         free_percpu(net->mib.xfrm_statistics);
3968 }
3969 #else
3970 static int __net_init xfrm_statistics_init(struct net *net)
3971 {
3972         return 0;
3973 }
3974
3975 static void xfrm_statistics_fini(struct net *net)
3976 {
3977 }
3978 #endif
3979
3980 static int __net_init xfrm_policy_init(struct net *net)
3981 {
3982         unsigned int hmask, sz;
3983         int dir, err;
3984
3985         if (net_eq(net, &init_net)) {
3986                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
3987                                            sizeof(struct xfrm_dst),
3988                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3989                                            NULL);
3990                 err = rhashtable_init(&xfrm_policy_inexact_table,
3991                                       &xfrm_pol_inexact_params);
3992                 BUG_ON(err);
3993         }
3994
3995         hmask = 8 - 1;
3996         sz = (hmask+1) * sizeof(struct hlist_head);
3997
3998         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
3999         if (!net->xfrm.policy_byidx)
4000                 goto out_byidx;
4001         net->xfrm.policy_idx_hmask = hmask;
4002
4003         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4004                 struct xfrm_policy_hash *htab;
4005
4006                 net->xfrm.policy_count[dir] = 0;
4007                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4008                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4009
4010                 htab = &net->xfrm.policy_bydst[dir];
4011                 htab->table = xfrm_hash_alloc(sz);
4012                 if (!htab->table)
4013                         goto out_bydst;
4014                 htab->hmask = hmask;
4015                 htab->dbits4 = 32;
4016                 htab->sbits4 = 32;
4017                 htab->dbits6 = 128;
4018                 htab->sbits6 = 128;
4019         }
4020         net->xfrm.policy_hthresh.lbits4 = 32;
4021         net->xfrm.policy_hthresh.rbits4 = 32;
4022         net->xfrm.policy_hthresh.lbits6 = 128;
4023         net->xfrm.policy_hthresh.rbits6 = 128;
4024
4025         seqlock_init(&net->xfrm.policy_hthresh.lock);
4026
4027         INIT_LIST_HEAD(&net->xfrm.policy_all);
4028         INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4029         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4030         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4031         return 0;
4032
4033 out_bydst:
4034         for (dir--; dir >= 0; dir--) {
4035                 struct xfrm_policy_hash *htab;
4036
4037                 htab = &net->xfrm.policy_bydst[dir];
4038                 xfrm_hash_free(htab->table, sz);
4039         }
4040         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4041 out_byidx:
4042         return -ENOMEM;
4043 }
4044
4045 static void xfrm_policy_fini(struct net *net)
4046 {
4047         struct xfrm_pol_inexact_bin *b, *t;
4048         unsigned int sz;
4049         int dir;
4050
4051         flush_work(&net->xfrm.policy_hash_work);
4052 #ifdef CONFIG_XFRM_SUB_POLICY
4053         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4054 #endif
4055         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4056
4057         WARN_ON(!list_empty(&net->xfrm.policy_all));
4058
4059         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4060                 struct xfrm_policy_hash *htab;
4061
4062                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4063
4064                 htab = &net->xfrm.policy_bydst[dir];
4065                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4066                 WARN_ON(!hlist_empty(htab->table));
4067                 xfrm_hash_free(htab->table, sz);
4068         }
4069
4070         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4071         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4072         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4073
4074         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4075         list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4076                 __xfrm_policy_inexact_prune_bin(b, true);
4077         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4078 }
4079
4080 static int __net_init xfrm_net_init(struct net *net)
4081 {
4082         int rv;
4083
4084         /* Initialize the per-net locks here */
4085         spin_lock_init(&net->xfrm.xfrm_state_lock);
4086         spin_lock_init(&net->xfrm.xfrm_policy_lock);
4087         mutex_init(&net->xfrm.xfrm_cfg_mutex);
4088
4089         rv = xfrm_statistics_init(net);
4090         if (rv < 0)
4091                 goto out_statistics;
4092         rv = xfrm_state_init(net);
4093         if (rv < 0)
4094                 goto out_state;
4095         rv = xfrm_policy_init(net);
4096         if (rv < 0)
4097                 goto out_policy;
4098         rv = xfrm_sysctl_init(net);
4099         if (rv < 0)
4100                 goto out_sysctl;
4101
4102         return 0;
4103
4104 out_sysctl:
4105         xfrm_policy_fini(net);
4106 out_policy:
4107         xfrm_state_fini(net);
4108 out_state:
4109         xfrm_statistics_fini(net);
4110 out_statistics:
4111         return rv;
4112 }
4113
4114 static void __net_exit xfrm_net_exit(struct net *net)
4115 {
4116         xfrm_sysctl_fini(net);
4117         xfrm_policy_fini(net);
4118         xfrm_state_fini(net);
4119         xfrm_statistics_fini(net);
4120 }
4121
4122 static struct pernet_operations __net_initdata xfrm_net_ops = {
4123         .init = xfrm_net_init,
4124         .exit = xfrm_net_exit,
4125 };
4126
4127 void __init xfrm_init(void)
4128 {
4129         register_pernet_subsys(&xfrm_net_ops);
4130         xfrm_dev_init();
4131         seqcount_mutex_init(&xfrm_policy_hash_generation, &hash_resize_mutex);
4132         xfrm_input_init();
4133
4134 #ifdef CONFIG_XFRM_ESPINTCP
4135         espintcp_init();
4136 #endif
4137 }
4138
4139 #ifdef CONFIG_AUDITSYSCALL
4140 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4141                                          struct audit_buffer *audit_buf)
4142 {
4143         struct xfrm_sec_ctx *ctx = xp->security;
4144         struct xfrm_selector *sel = &xp->selector;
4145
4146         if (ctx)
4147                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4148                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4149
4150         switch (sel->family) {
4151         case AF_INET:
4152                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4153                 if (sel->prefixlen_s != 32)
4154                         audit_log_format(audit_buf, " src_prefixlen=%d",
4155                                          sel->prefixlen_s);
4156                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4157                 if (sel->prefixlen_d != 32)
4158                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4159                                          sel->prefixlen_d);
4160                 break;
4161         case AF_INET6:
4162                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4163                 if (sel->prefixlen_s != 128)
4164                         audit_log_format(audit_buf, " src_prefixlen=%d",
4165                                          sel->prefixlen_s);
4166                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4167                 if (sel->prefixlen_d != 128)
4168                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4169                                          sel->prefixlen_d);
4170                 break;
4171         }
4172 }
4173
4174 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4175 {
4176         struct audit_buffer *audit_buf;
4177
4178         audit_buf = xfrm_audit_start("SPD-add");
4179         if (audit_buf == NULL)
4180                 return;
4181         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4182         audit_log_format(audit_buf, " res=%u", result);
4183         xfrm_audit_common_policyinfo(xp, audit_buf);
4184         audit_log_end(audit_buf);
4185 }
4186 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4187
4188 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4189                               bool task_valid)
4190 {
4191         struct audit_buffer *audit_buf;
4192
4193         audit_buf = xfrm_audit_start("SPD-delete");
4194         if (audit_buf == NULL)
4195                 return;
4196         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4197         audit_log_format(audit_buf, " res=%u", result);
4198         xfrm_audit_common_policyinfo(xp, audit_buf);
4199         audit_log_end(audit_buf);
4200 }
4201 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4202 #endif
4203
4204 #ifdef CONFIG_XFRM_MIGRATE
4205 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4206                                         const struct xfrm_selector *sel_tgt)
4207 {
4208         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4209                 if (sel_tgt->family == sel_cmp->family &&
4210                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4211                                     sel_cmp->family) &&
4212                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4213                                     sel_cmp->family) &&
4214                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4215                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4216                         return true;
4217                 }
4218         } else {
4219                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4220                         return true;
4221                 }
4222         }
4223         return false;
4224 }
4225
4226 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4227                                                     u8 dir, u8 type, struct net *net)
4228 {
4229         struct xfrm_policy *pol, *ret = NULL;
4230         struct hlist_head *chain;
4231         u32 priority = ~0U;
4232
4233         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4234         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4235         hlist_for_each_entry(pol, chain, bydst) {
4236                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4237                     pol->type == type) {
4238                         ret = pol;
4239                         priority = ret->priority;
4240                         break;
4241                 }
4242         }
4243         chain = &net->xfrm.policy_inexact[dir];
4244         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4245                 if ((pol->priority >= priority) && ret)
4246                         break;
4247
4248                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4249                     pol->type == type) {
4250                         ret = pol;
4251                         break;
4252                 }
4253         }
4254
4255         xfrm_pol_hold(ret);
4256
4257         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4258
4259         return ret;
4260 }
4261
4262 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4263 {
4264         int match = 0;
4265
4266         if (t->mode == m->mode && t->id.proto == m->proto &&
4267             (m->reqid == 0 || t->reqid == m->reqid)) {
4268                 switch (t->mode) {
4269                 case XFRM_MODE_TUNNEL:
4270                 case XFRM_MODE_BEET:
4271                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4272                                             m->old_family) &&
4273                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
4274                                             m->old_family)) {
4275                                 match = 1;
4276                         }
4277                         break;
4278                 case XFRM_MODE_TRANSPORT:
4279                         /* in case of transport mode, template does not store
4280                            any IP addresses, hence we just compare mode and
4281                            protocol */
4282                         match = 1;
4283                         break;
4284                 default:
4285                         break;
4286                 }
4287         }
4288         return match;
4289 }
4290
4291 /* update endpoint address(es) of template(s) */
4292 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4293                                struct xfrm_migrate *m, int num_migrate)
4294 {
4295         struct xfrm_migrate *mp;
4296         int i, j, n = 0;
4297
4298         write_lock_bh(&pol->lock);
4299         if (unlikely(pol->walk.dead)) {
4300                 /* target policy has been deleted */
4301                 write_unlock_bh(&pol->lock);
4302                 return -ENOENT;
4303         }
4304
4305         for (i = 0; i < pol->xfrm_nr; i++) {
4306                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4307                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4308                                 continue;
4309                         n++;
4310                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4311                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4312                                 continue;
4313                         /* update endpoints */
4314                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4315                                sizeof(pol->xfrm_vec[i].id.daddr));
4316                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4317                                sizeof(pol->xfrm_vec[i].saddr));
4318                         pol->xfrm_vec[i].encap_family = mp->new_family;
4319                         /* flush bundles */
4320                         atomic_inc(&pol->genid);
4321                 }
4322         }
4323
4324         write_unlock_bh(&pol->lock);
4325
4326         if (!n)
4327                 return -ENODATA;
4328
4329         return 0;
4330 }
4331
4332 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4333 {
4334         int i, j;
4335
4336         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4337                 return -EINVAL;
4338
4339         for (i = 0; i < num_migrate; i++) {
4340                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4341                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4342                         return -EINVAL;
4343
4344                 /* check if there is any duplicated entry */
4345                 for (j = i + 1; j < num_migrate; j++) {
4346                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4347                                     sizeof(m[i].old_daddr)) &&
4348                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4349                                     sizeof(m[i].old_saddr)) &&
4350                             m[i].proto == m[j].proto &&
4351                             m[i].mode == m[j].mode &&
4352                             m[i].reqid == m[j].reqid &&
4353                             m[i].old_family == m[j].old_family)
4354                                 return -EINVAL;
4355                 }
4356         }
4357
4358         return 0;
4359 }
4360
4361 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4362                  struct xfrm_migrate *m, int num_migrate,
4363                  struct xfrm_kmaddress *k, struct net *net,
4364                  struct xfrm_encap_tmpl *encap)
4365 {
4366         int i, err, nx_cur = 0, nx_new = 0;
4367         struct xfrm_policy *pol = NULL;
4368         struct xfrm_state *x, *xc;
4369         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4370         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4371         struct xfrm_migrate *mp;
4372
4373         /* Stage 0 - sanity checks */
4374         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4375                 goto out;
4376
4377         if (dir >= XFRM_POLICY_MAX) {
4378                 err = -EINVAL;
4379                 goto out;
4380         }
4381
4382         /* Stage 1 - find policy */
4383         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4384                 err = -ENOENT;
4385                 goto out;
4386         }
4387
4388         /* Stage 2 - find and update state(s) */
4389         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4390                 if ((x = xfrm_migrate_state_find(mp, net))) {
4391                         x_cur[nx_cur] = x;
4392                         nx_cur++;
4393                         xc = xfrm_state_migrate(x, mp, encap);
4394                         if (xc) {
4395                                 x_new[nx_new] = xc;
4396                                 nx_new++;
4397                         } else {
4398                                 err = -ENODATA;
4399                                 goto restore_state;
4400                         }
4401                 }
4402         }
4403
4404         /* Stage 3 - update policy */
4405         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4406                 goto restore_state;
4407
4408         /* Stage 4 - delete old state(s) */
4409         if (nx_cur) {
4410                 xfrm_states_put(x_cur, nx_cur);
4411                 xfrm_states_delete(x_cur, nx_cur);
4412         }
4413
4414         /* Stage 5 - announce */
4415         km_migrate(sel, dir, type, m, num_migrate, k, encap);
4416
4417         xfrm_pol_put(pol);
4418
4419         return 0;
4420 out:
4421         return err;
4422
4423 restore_state:
4424         if (pol)
4425                 xfrm_pol_put(pol);
4426         if (nx_cur)
4427                 xfrm_states_put(x_cur, nx_cur);
4428         if (nx_new)
4429                 xfrm_states_delete(x_new, nx_new);
4430
4431         return err;
4432 }
4433 EXPORT_SYMBOL(xfrm_migrate);
4434 #endif