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         return ret;
1907 }
1908
1909 static struct xfrm_pol_inexact_node *
1910 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1911                                 seqcount_spinlock_t *count,
1912                                 const xfrm_address_t *addr, u16 family)
1913 {
1914         const struct rb_node *parent;
1915         int seq;
1916
1917 again:
1918         seq = read_seqcount_begin(count);
1919
1920         parent = rcu_dereference_raw(r->rb_node);
1921         while (parent) {
1922                 struct xfrm_pol_inexact_node *node;
1923                 int delta;
1924
1925                 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1926
1927                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1928                                                node->prefixlen, family);
1929                 if (delta < 0) {
1930                         parent = rcu_dereference_raw(parent->rb_left);
1931                         continue;
1932                 } else if (delta > 0) {
1933                         parent = rcu_dereference_raw(parent->rb_right);
1934                         continue;
1935                 }
1936
1937                 return node;
1938         }
1939
1940         if (read_seqcount_retry(count, seq))
1941                 goto again;
1942
1943         return NULL;
1944 }
1945
1946 static bool
1947 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1948                                     struct xfrm_pol_inexact_bin *b,
1949                                     const xfrm_address_t *saddr,
1950                                     const xfrm_address_t *daddr)
1951 {
1952         struct xfrm_pol_inexact_node *n;
1953         u16 family;
1954
1955         if (!b)
1956                 return false;
1957
1958         family = b->k.family;
1959         memset(cand, 0, sizeof(*cand));
1960         cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1961
1962         n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1963                                             family);
1964         if (n) {
1965                 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1966                 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1967                                                     family);
1968                 if (n)
1969                         cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1970         }
1971
1972         n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1973                                             family);
1974         if (n)
1975                 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1976
1977         return true;
1978 }
1979
1980 static struct xfrm_pol_inexact_bin *
1981 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1982                                u8 dir, u32 if_id)
1983 {
1984         struct xfrm_pol_inexact_key k = {
1985                 .family = family,
1986                 .type = type,
1987                 .dir = dir,
1988                 .if_id = if_id,
1989         };
1990
1991         write_pnet(&k.net, net);
1992
1993         return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1994                                  xfrm_pol_inexact_params);
1995 }
1996
1997 static struct xfrm_pol_inexact_bin *
1998 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1999                            u8 dir, u32 if_id)
2000 {
2001         struct xfrm_pol_inexact_bin *bin;
2002
2003         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2004
2005         rcu_read_lock();
2006         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2007         rcu_read_unlock();
2008
2009         return bin;
2010 }
2011
2012 static struct xfrm_policy *
2013 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2014                               struct xfrm_policy *prefer,
2015                               const struct flowi *fl,
2016                               u8 type, u16 family, int dir, u32 if_id)
2017 {
2018         u32 priority = prefer ? prefer->priority : ~0u;
2019         struct xfrm_policy *pol;
2020
2021         if (!chain)
2022                 return NULL;
2023
2024         hlist_for_each_entry_rcu(pol, chain, bydst) {
2025                 int err;
2026
2027                 if (pol->priority > priority)
2028                         break;
2029
2030                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2031                 if (err) {
2032                         if (err != -ESRCH)
2033                                 return ERR_PTR(err);
2034
2035                         continue;
2036                 }
2037
2038                 if (prefer) {
2039                         /* matches.  Is it older than *prefer? */
2040                         if (pol->priority == priority &&
2041                             prefer->pos < pol->pos)
2042                                 return prefer;
2043                 }
2044
2045                 return pol;
2046         }
2047
2048         return NULL;
2049 }
2050
2051 static struct xfrm_policy *
2052 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2053                             struct xfrm_policy *prefer,
2054                             const struct flowi *fl,
2055                             u8 type, u16 family, int dir, u32 if_id)
2056 {
2057         struct xfrm_policy *tmp;
2058         int i;
2059
2060         for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2061                 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2062                                                     prefer,
2063                                                     fl, type, family, dir,
2064                                                     if_id);
2065                 if (!tmp)
2066                         continue;
2067
2068                 if (IS_ERR(tmp))
2069                         return tmp;
2070                 prefer = tmp;
2071         }
2072
2073         return prefer;
2074 }
2075
2076 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2077                                                      const struct flowi *fl,
2078                                                      u16 family, u8 dir,
2079                                                      u32 if_id)
2080 {
2081         struct xfrm_pol_inexact_candidates cand;
2082         const xfrm_address_t *daddr, *saddr;
2083         struct xfrm_pol_inexact_bin *bin;
2084         struct xfrm_policy *pol, *ret;
2085         struct hlist_head *chain;
2086         unsigned int sequence;
2087         int err;
2088
2089         daddr = xfrm_flowi_daddr(fl, family);
2090         saddr = xfrm_flowi_saddr(fl, family);
2091         if (unlikely(!daddr || !saddr))
2092                 return NULL;
2093
2094  retry:
2095         sequence = read_seqcount_begin(&xfrm_policy_hash_generation);
2096         rcu_read_lock();
2097
2098         chain = policy_hash_direct(net, daddr, saddr, family, dir);
2099         if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence)) {
2100                 rcu_read_unlock();
2101                 goto retry;
2102         }
2103
2104         ret = NULL;
2105         hlist_for_each_entry_rcu(pol, chain, bydst) {
2106                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2107                 if (err) {
2108                         if (err == -ESRCH)
2109                                 continue;
2110                         else {
2111                                 ret = ERR_PTR(err);
2112                                 goto fail;
2113                         }
2114                 } else {
2115                         ret = pol;
2116                         break;
2117                 }
2118         }
2119         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2120         if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2121                                                          daddr))
2122                 goto skip_inexact;
2123
2124         pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2125                                           family, dir, if_id);
2126         if (pol) {
2127                 ret = pol;
2128                 if (IS_ERR(pol))
2129                         goto fail;
2130         }
2131
2132 skip_inexact:
2133         if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence)) {
2134                 rcu_read_unlock();
2135                 goto retry;
2136         }
2137
2138         if (ret && !xfrm_pol_hold_rcu(ret)) {
2139                 rcu_read_unlock();
2140                 goto retry;
2141         }
2142 fail:
2143         rcu_read_unlock();
2144
2145         return ret;
2146 }
2147
2148 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2149                                               const struct flowi *fl,
2150                                               u16 family, u8 dir, u32 if_id)
2151 {
2152 #ifdef CONFIG_XFRM_SUB_POLICY
2153         struct xfrm_policy *pol;
2154
2155         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2156                                         dir, if_id);
2157         if (pol != NULL)
2158                 return pol;
2159 #endif
2160         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2161                                          dir, if_id);
2162 }
2163
2164 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2165                                                  const struct flowi *fl,
2166                                                  u16 family, u32 if_id)
2167 {
2168         struct xfrm_policy *pol;
2169
2170         rcu_read_lock();
2171  again:
2172         pol = rcu_dereference(sk->sk_policy[dir]);
2173         if (pol != NULL) {
2174                 bool match;
2175                 int err = 0;
2176
2177                 if (pol->family != family) {
2178                         pol = NULL;
2179                         goto out;
2180                 }
2181
2182                 match = xfrm_selector_match(&pol->selector, fl, family);
2183                 if (match) {
2184                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2185                             pol->if_id != if_id) {
2186                                 pol = NULL;
2187                                 goto out;
2188                         }
2189                         err = security_xfrm_policy_lookup(pol->security,
2190                                                       fl->flowi_secid);
2191                         if (!err) {
2192                                 if (!xfrm_pol_hold_rcu(pol))
2193                                         goto again;
2194                         } else if (err == -ESRCH) {
2195                                 pol = NULL;
2196                         } else {
2197                                 pol = ERR_PTR(err);
2198                         }
2199                 } else
2200                         pol = NULL;
2201         }
2202 out:
2203         rcu_read_unlock();
2204         return pol;
2205 }
2206
2207 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2208 {
2209         struct net *net = xp_net(pol);
2210
2211         list_add(&pol->walk.all, &net->xfrm.policy_all);
2212         net->xfrm.policy_count[dir]++;
2213         xfrm_pol_hold(pol);
2214 }
2215
2216 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2217                                                 int dir)
2218 {
2219         struct net *net = xp_net(pol);
2220
2221         if (list_empty(&pol->walk.all))
2222                 return NULL;
2223
2224         /* Socket policies are not hashed. */
2225         if (!hlist_unhashed(&pol->bydst)) {
2226                 hlist_del_rcu(&pol->bydst);
2227                 hlist_del_init(&pol->bydst_inexact_list);
2228                 hlist_del(&pol->byidx);
2229         }
2230
2231         list_del_init(&pol->walk.all);
2232         net->xfrm.policy_count[dir]--;
2233
2234         return pol;
2235 }
2236
2237 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2238 {
2239         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2240 }
2241
2242 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2243 {
2244         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2245 }
2246
2247 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2248 {
2249         struct net *net = xp_net(pol);
2250
2251         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2252         pol = __xfrm_policy_unlink(pol, dir);
2253         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2254         if (pol) {
2255                 xfrm_policy_kill(pol);
2256                 return 0;
2257         }
2258         return -ENOENT;
2259 }
2260 EXPORT_SYMBOL(xfrm_policy_delete);
2261
2262 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2263 {
2264         struct net *net = sock_net(sk);
2265         struct xfrm_policy *old_pol;
2266
2267 #ifdef CONFIG_XFRM_SUB_POLICY
2268         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2269                 return -EINVAL;
2270 #endif
2271
2272         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2273         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2274                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2275         if (pol) {
2276                 pol->curlft.add_time = ktime_get_real_seconds();
2277                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2278                 xfrm_sk_policy_link(pol, dir);
2279         }
2280         rcu_assign_pointer(sk->sk_policy[dir], pol);
2281         if (old_pol) {
2282                 if (pol)
2283                         xfrm_policy_requeue(old_pol, pol);
2284
2285                 /* Unlinking succeeds always. This is the only function
2286                  * allowed to delete or replace socket policy.
2287                  */
2288                 xfrm_sk_policy_unlink(old_pol, dir);
2289         }
2290         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2291
2292         if (old_pol) {
2293                 xfrm_policy_kill(old_pol);
2294         }
2295         return 0;
2296 }
2297
2298 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2299 {
2300         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2301         struct net *net = xp_net(old);
2302
2303         if (newp) {
2304                 newp->selector = old->selector;
2305                 if (security_xfrm_policy_clone(old->security,
2306                                                &newp->security)) {
2307                         kfree(newp);
2308                         return NULL;  /* ENOMEM */
2309                 }
2310                 newp->lft = old->lft;
2311                 newp->curlft = old->curlft;
2312                 newp->mark = old->mark;
2313                 newp->if_id = old->if_id;
2314                 newp->action = old->action;
2315                 newp->flags = old->flags;
2316                 newp->xfrm_nr = old->xfrm_nr;
2317                 newp->index = old->index;
2318                 newp->type = old->type;
2319                 newp->family = old->family;
2320                 memcpy(newp->xfrm_vec, old->xfrm_vec,
2321                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2322                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2323                 xfrm_sk_policy_link(newp, dir);
2324                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2325                 xfrm_pol_put(newp);
2326         }
2327         return newp;
2328 }
2329
2330 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2331 {
2332         const struct xfrm_policy *p;
2333         struct xfrm_policy *np;
2334         int i, ret = 0;
2335
2336         rcu_read_lock();
2337         for (i = 0; i < 2; i++) {
2338                 p = rcu_dereference(osk->sk_policy[i]);
2339                 if (p) {
2340                         np = clone_policy(p, i);
2341                         if (unlikely(!np)) {
2342                                 ret = -ENOMEM;
2343                                 break;
2344                         }
2345                         rcu_assign_pointer(sk->sk_policy[i], np);
2346                 }
2347         }
2348         rcu_read_unlock();
2349         return ret;
2350 }
2351
2352 static int
2353 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2354                xfrm_address_t *remote, unsigned short family, u32 mark)
2355 {
2356         int err;
2357         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2358
2359         if (unlikely(afinfo == NULL))
2360                 return -EINVAL;
2361         err = afinfo->get_saddr(net, oif, local, remote, mark);
2362         rcu_read_unlock();
2363         return err;
2364 }
2365
2366 /* Resolve list of templates for the flow, given policy. */
2367
2368 static int
2369 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2370                       struct xfrm_state **xfrm, unsigned short family)
2371 {
2372         struct net *net = xp_net(policy);
2373         int nx;
2374         int i, error;
2375         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2376         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2377         xfrm_address_t tmp;
2378
2379         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2380                 struct xfrm_state *x;
2381                 xfrm_address_t *remote = daddr;
2382                 xfrm_address_t *local  = saddr;
2383                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2384
2385                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2386                     tmpl->mode == XFRM_MODE_BEET) {
2387                         remote = &tmpl->id.daddr;
2388                         local = &tmpl->saddr;
2389                         if (xfrm_addr_any(local, tmpl->encap_family)) {
2390                                 error = xfrm_get_saddr(net, fl->flowi_oif,
2391                                                        &tmp, remote,
2392                                                        tmpl->encap_family, 0);
2393                                 if (error)
2394                                         goto fail;
2395                                 local = &tmp;
2396                         }
2397                 }
2398
2399                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2400                                     family, policy->if_id);
2401
2402                 if (x && x->km.state == XFRM_STATE_VALID) {
2403                         xfrm[nx++] = x;
2404                         daddr = remote;
2405                         saddr = local;
2406                         continue;
2407                 }
2408                 if (x) {
2409                         error = (x->km.state == XFRM_STATE_ERROR ?
2410                                  -EINVAL : -EAGAIN);
2411                         xfrm_state_put(x);
2412                 } else if (error == -ESRCH) {
2413                         error = -EAGAIN;
2414                 }
2415
2416                 if (!tmpl->optional)
2417                         goto fail;
2418         }
2419         return nx;
2420
2421 fail:
2422         for (nx--; nx >= 0; nx--)
2423                 xfrm_state_put(xfrm[nx]);
2424         return error;
2425 }
2426
2427 static int
2428 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2429                   struct xfrm_state **xfrm, unsigned short family)
2430 {
2431         struct xfrm_state *tp[XFRM_MAX_DEPTH];
2432         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2433         int cnx = 0;
2434         int error;
2435         int ret;
2436         int i;
2437
2438         for (i = 0; i < npols; i++) {
2439                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2440                         error = -ENOBUFS;
2441                         goto fail;
2442                 }
2443
2444                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2445                 if (ret < 0) {
2446                         error = ret;
2447                         goto fail;
2448                 } else
2449                         cnx += ret;
2450         }
2451
2452         /* found states are sorted for outbound processing */
2453         if (npols > 1)
2454                 xfrm_state_sort(xfrm, tpp, cnx, family);
2455
2456         return cnx;
2457
2458  fail:
2459         for (cnx--; cnx >= 0; cnx--)
2460                 xfrm_state_put(tpp[cnx]);
2461         return error;
2462
2463 }
2464
2465 static int xfrm_get_tos(const struct flowi *fl, int family)
2466 {
2467         if (family == AF_INET)
2468                 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2469
2470         return 0;
2471 }
2472
2473 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2474 {
2475         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2476         struct dst_ops *dst_ops;
2477         struct xfrm_dst *xdst;
2478
2479         if (!afinfo)
2480                 return ERR_PTR(-EINVAL);
2481
2482         switch (family) {
2483         case AF_INET:
2484                 dst_ops = &net->xfrm.xfrm4_dst_ops;
2485                 break;
2486 #if IS_ENABLED(CONFIG_IPV6)
2487         case AF_INET6:
2488                 dst_ops = &net->xfrm.xfrm6_dst_ops;
2489                 break;
2490 #endif
2491         default:
2492                 BUG();
2493         }
2494         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2495
2496         if (likely(xdst)) {
2497                 struct dst_entry *dst = &xdst->u.dst;
2498
2499                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2500         } else
2501                 xdst = ERR_PTR(-ENOBUFS);
2502
2503         rcu_read_unlock();
2504
2505         return xdst;
2506 }
2507
2508 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2509                            int nfheader_len)
2510 {
2511         if (dst->ops->family == AF_INET6) {
2512                 struct rt6_info *rt = (struct rt6_info *)dst;
2513                 path->path_cookie = rt6_get_cookie(rt);
2514                 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2515         }
2516 }
2517
2518 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2519                                 const struct flowi *fl)
2520 {
2521         const struct xfrm_policy_afinfo *afinfo =
2522                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2523         int err;
2524
2525         if (!afinfo)
2526                 return -EINVAL;
2527
2528         err = afinfo->fill_dst(xdst, dev, fl);
2529
2530         rcu_read_unlock();
2531
2532         return err;
2533 }
2534
2535
2536 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2537  * all the metrics... Shortly, bundle a bundle.
2538  */
2539
2540 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2541                                             struct xfrm_state **xfrm,
2542                                             struct xfrm_dst **bundle,
2543                                             int nx,
2544                                             const struct flowi *fl,
2545                                             struct dst_entry *dst)
2546 {
2547         const struct xfrm_state_afinfo *afinfo;
2548         const struct xfrm_mode *inner_mode;
2549         struct net *net = xp_net(policy);
2550         unsigned long now = jiffies;
2551         struct net_device *dev;
2552         struct xfrm_dst *xdst_prev = NULL;
2553         struct xfrm_dst *xdst0 = NULL;
2554         int i = 0;
2555         int err;
2556         int header_len = 0;
2557         int nfheader_len = 0;
2558         int trailer_len = 0;
2559         int tos;
2560         int family = policy->selector.family;
2561         xfrm_address_t saddr, daddr;
2562
2563         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2564
2565         tos = xfrm_get_tos(fl, family);
2566
2567         dst_hold(dst);
2568
2569         for (; i < nx; i++) {
2570                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2571                 struct dst_entry *dst1 = &xdst->u.dst;
2572
2573                 err = PTR_ERR(xdst);
2574                 if (IS_ERR(xdst)) {
2575                         dst_release(dst);
2576                         goto put_states;
2577                 }
2578
2579                 bundle[i] = xdst;
2580                 if (!xdst_prev)
2581                         xdst0 = xdst;
2582                 else
2583                         /* Ref count is taken during xfrm_alloc_dst()
2584                          * No need to do dst_clone() on dst1
2585                          */
2586                         xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2587
2588                 if (xfrm[i]->sel.family == AF_UNSPEC) {
2589                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
2590                                                         xfrm_af2proto(family));
2591                         if (!inner_mode) {
2592                                 err = -EAFNOSUPPORT;
2593                                 dst_release(dst);
2594                                 goto put_states;
2595                         }
2596                 } else
2597                         inner_mode = &xfrm[i]->inner_mode;
2598
2599                 xdst->route = dst;
2600                 dst_copy_metrics(dst1, dst);
2601
2602                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2603                         __u32 mark = 0;
2604
2605                         if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2606                                 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2607
2608                         family = xfrm[i]->props.family;
2609                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2610                                               &saddr, &daddr, family, mark);
2611                         err = PTR_ERR(dst);
2612                         if (IS_ERR(dst))
2613                                 goto put_states;
2614                 } else
2615                         dst_hold(dst);
2616
2617                 dst1->xfrm = xfrm[i];
2618                 xdst->xfrm_genid = xfrm[i]->genid;
2619
2620                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2621                 dst1->lastuse = now;
2622
2623                 dst1->input = dst_discard;
2624
2625                 rcu_read_lock();
2626                 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2627                 if (likely(afinfo))
2628                         dst1->output = afinfo->output;
2629                 else
2630                         dst1->output = dst_discard_out;
2631                 rcu_read_unlock();
2632
2633                 xdst_prev = xdst;
2634
2635                 header_len += xfrm[i]->props.header_len;
2636                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2637                         nfheader_len += xfrm[i]->props.header_len;
2638                 trailer_len += xfrm[i]->props.trailer_len;
2639         }
2640
2641         xfrm_dst_set_child(xdst_prev, dst);
2642         xdst0->path = dst;
2643
2644         err = -ENODEV;
2645         dev = dst->dev;
2646         if (!dev)
2647                 goto free_dst;
2648
2649         xfrm_init_path(xdst0, dst, nfheader_len);
2650         xfrm_init_pmtu(bundle, nx);
2651
2652         for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2653              xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2654                 err = xfrm_fill_dst(xdst_prev, dev, fl);
2655                 if (err)
2656                         goto free_dst;
2657
2658                 xdst_prev->u.dst.header_len = header_len;
2659                 xdst_prev->u.dst.trailer_len = trailer_len;
2660                 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2661                 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2662         }
2663
2664         return &xdst0->u.dst;
2665
2666 put_states:
2667         for (; i < nx; i++)
2668                 xfrm_state_put(xfrm[i]);
2669 free_dst:
2670         if (xdst0)
2671                 dst_release_immediate(&xdst0->u.dst);
2672
2673         return ERR_PTR(err);
2674 }
2675
2676 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2677                                 struct xfrm_policy **pols,
2678                                 int *num_pols, int *num_xfrms)
2679 {
2680         int i;
2681
2682         if (*num_pols == 0 || !pols[0]) {
2683                 *num_pols = 0;
2684                 *num_xfrms = 0;
2685                 return 0;
2686         }
2687         if (IS_ERR(pols[0]))
2688                 return PTR_ERR(pols[0]);
2689
2690         *num_xfrms = pols[0]->xfrm_nr;
2691
2692 #ifdef CONFIG_XFRM_SUB_POLICY
2693         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2694             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2695                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2696                                                     XFRM_POLICY_TYPE_MAIN,
2697                                                     fl, family,
2698                                                     XFRM_POLICY_OUT,
2699                                                     pols[0]->if_id);
2700                 if (pols[1]) {
2701                         if (IS_ERR(pols[1])) {
2702                                 xfrm_pols_put(pols, *num_pols);
2703                                 return PTR_ERR(pols[1]);
2704                         }
2705                         (*num_pols)++;
2706                         (*num_xfrms) += pols[1]->xfrm_nr;
2707                 }
2708         }
2709 #endif
2710         for (i = 0; i < *num_pols; i++) {
2711                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2712                         *num_xfrms = -1;
2713                         break;
2714                 }
2715         }
2716
2717         return 0;
2718
2719 }
2720
2721 static struct xfrm_dst *
2722 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2723                                const struct flowi *fl, u16 family,
2724                                struct dst_entry *dst_orig)
2725 {
2726         struct net *net = xp_net(pols[0]);
2727         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2728         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2729         struct xfrm_dst *xdst;
2730         struct dst_entry *dst;
2731         int err;
2732
2733         /* Try to instantiate a bundle */
2734         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2735         if (err <= 0) {
2736                 if (err == 0)
2737                         return NULL;
2738
2739                 if (err != -EAGAIN)
2740                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2741                 return ERR_PTR(err);
2742         }
2743
2744         dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2745         if (IS_ERR(dst)) {
2746                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2747                 return ERR_CAST(dst);
2748         }
2749
2750         xdst = (struct xfrm_dst *)dst;
2751         xdst->num_xfrms = err;
2752         xdst->num_pols = num_pols;
2753         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2754         xdst->policy_genid = atomic_read(&pols[0]->genid);
2755
2756         return xdst;
2757 }
2758
2759 static void xfrm_policy_queue_process(struct timer_list *t)
2760 {
2761         struct sk_buff *skb;
2762         struct sock *sk;
2763         struct dst_entry *dst;
2764         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2765         struct net *net = xp_net(pol);
2766         struct xfrm_policy_queue *pq = &pol->polq;
2767         struct flowi fl;
2768         struct sk_buff_head list;
2769         __u32 skb_mark;
2770
2771         spin_lock(&pq->hold_queue.lock);
2772         skb = skb_peek(&pq->hold_queue);
2773         if (!skb) {
2774                 spin_unlock(&pq->hold_queue.lock);
2775                 goto out;
2776         }
2777         dst = skb_dst(skb);
2778         sk = skb->sk;
2779
2780         /* Fixup the mark to support VTI. */
2781         skb_mark = skb->mark;
2782         skb->mark = pol->mark.v;
2783         xfrm_decode_session(skb, &fl, dst->ops->family);
2784         skb->mark = skb_mark;
2785         spin_unlock(&pq->hold_queue.lock);
2786
2787         dst_hold(xfrm_dst_path(dst));
2788         dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2789         if (IS_ERR(dst))
2790                 goto purge_queue;
2791
2792         if (dst->flags & DST_XFRM_QUEUE) {
2793                 dst_release(dst);
2794
2795                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2796                         goto purge_queue;
2797
2798                 pq->timeout = pq->timeout << 1;
2799                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2800                         xfrm_pol_hold(pol);
2801                 goto out;
2802         }
2803
2804         dst_release(dst);
2805
2806         __skb_queue_head_init(&list);
2807
2808         spin_lock(&pq->hold_queue.lock);
2809         pq->timeout = 0;
2810         skb_queue_splice_init(&pq->hold_queue, &list);
2811         spin_unlock(&pq->hold_queue.lock);
2812
2813         while (!skb_queue_empty(&list)) {
2814                 skb = __skb_dequeue(&list);
2815
2816                 /* Fixup the mark to support VTI. */
2817                 skb_mark = skb->mark;
2818                 skb->mark = pol->mark.v;
2819                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2820                 skb->mark = skb_mark;
2821
2822                 dst_hold(xfrm_dst_path(skb_dst(skb)));
2823                 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2824                 if (IS_ERR(dst)) {
2825                         kfree_skb(skb);
2826                         continue;
2827                 }
2828
2829                 nf_reset_ct(skb);
2830                 skb_dst_drop(skb);
2831                 skb_dst_set(skb, dst);
2832
2833                 dst_output(net, skb->sk, skb);
2834         }
2835
2836 out:
2837         xfrm_pol_put(pol);
2838         return;
2839
2840 purge_queue:
2841         pq->timeout = 0;
2842         skb_queue_purge(&pq->hold_queue);
2843         xfrm_pol_put(pol);
2844 }
2845
2846 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2847 {
2848         unsigned long sched_next;
2849         struct dst_entry *dst = skb_dst(skb);
2850         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2851         struct xfrm_policy *pol = xdst->pols[0];
2852         struct xfrm_policy_queue *pq = &pol->polq;
2853
2854         if (unlikely(skb_fclone_busy(sk, skb))) {
2855                 kfree_skb(skb);
2856                 return 0;
2857         }
2858
2859         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2860                 kfree_skb(skb);
2861                 return -EAGAIN;
2862         }
2863
2864         skb_dst_force(skb);
2865
2866         spin_lock_bh(&pq->hold_queue.lock);
2867
2868         if (!pq->timeout)
2869                 pq->timeout = XFRM_QUEUE_TMO_MIN;
2870
2871         sched_next = jiffies + pq->timeout;
2872
2873         if (del_timer(&pq->hold_timer)) {
2874                 if (time_before(pq->hold_timer.expires, sched_next))
2875                         sched_next = pq->hold_timer.expires;
2876                 xfrm_pol_put(pol);
2877         }
2878
2879         __skb_queue_tail(&pq->hold_queue, skb);
2880         if (!mod_timer(&pq->hold_timer, sched_next))
2881                 xfrm_pol_hold(pol);
2882
2883         spin_unlock_bh(&pq->hold_queue.lock);
2884
2885         return 0;
2886 }
2887
2888 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2889                                                  struct xfrm_flo *xflo,
2890                                                  const struct flowi *fl,
2891                                                  int num_xfrms,
2892                                                  u16 family)
2893 {
2894         int err;
2895         struct net_device *dev;
2896         struct dst_entry *dst;
2897         struct dst_entry *dst1;
2898         struct xfrm_dst *xdst;
2899
2900         xdst = xfrm_alloc_dst(net, family);
2901         if (IS_ERR(xdst))
2902                 return xdst;
2903
2904         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2905             net->xfrm.sysctl_larval_drop ||
2906             num_xfrms <= 0)
2907                 return xdst;
2908
2909         dst = xflo->dst_orig;
2910         dst1 = &xdst->u.dst;
2911         dst_hold(dst);
2912         xdst->route = dst;
2913
2914         dst_copy_metrics(dst1, dst);
2915
2916         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2917         dst1->flags |= DST_XFRM_QUEUE;
2918         dst1->lastuse = jiffies;
2919
2920         dst1->input = dst_discard;
2921         dst1->output = xdst_queue_output;
2922
2923         dst_hold(dst);
2924         xfrm_dst_set_child(xdst, dst);
2925         xdst->path = dst;
2926
2927         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2928
2929         err = -ENODEV;
2930         dev = dst->dev;
2931         if (!dev)
2932                 goto free_dst;
2933
2934         err = xfrm_fill_dst(xdst, dev, fl);
2935         if (err)
2936                 goto free_dst;
2937
2938 out:
2939         return xdst;
2940
2941 free_dst:
2942         dst_release(dst1);
2943         xdst = ERR_PTR(err);
2944         goto out;
2945 }
2946
2947 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2948                                            const struct flowi *fl,
2949                                            u16 family, u8 dir,
2950                                            struct xfrm_flo *xflo, u32 if_id)
2951 {
2952         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2953         int num_pols = 0, num_xfrms = 0, err;
2954         struct xfrm_dst *xdst;
2955
2956         /* Resolve policies to use if we couldn't get them from
2957          * previous cache entry */
2958         num_pols = 1;
2959         pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2960         err = xfrm_expand_policies(fl, family, pols,
2961                                            &num_pols, &num_xfrms);
2962         if (err < 0)
2963                 goto inc_error;
2964         if (num_pols == 0)
2965                 return NULL;
2966         if (num_xfrms <= 0)
2967                 goto make_dummy_bundle;
2968
2969         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2970                                               xflo->dst_orig);
2971         if (IS_ERR(xdst)) {
2972                 err = PTR_ERR(xdst);
2973                 if (err == -EREMOTE) {
2974                         xfrm_pols_put(pols, num_pols);
2975                         return NULL;
2976                 }
2977
2978                 if (err != -EAGAIN)
2979                         goto error;
2980                 goto make_dummy_bundle;
2981         } else if (xdst == NULL) {
2982                 num_xfrms = 0;
2983                 goto make_dummy_bundle;
2984         }
2985
2986         return xdst;
2987
2988 make_dummy_bundle:
2989         /* We found policies, but there's no bundles to instantiate:
2990          * either because the policy blocks, has no transformations or
2991          * we could not build template (no xfrm_states).*/
2992         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2993         if (IS_ERR(xdst)) {
2994                 xfrm_pols_put(pols, num_pols);
2995                 return ERR_CAST(xdst);
2996         }
2997         xdst->num_pols = num_pols;
2998         xdst->num_xfrms = num_xfrms;
2999         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
3000
3001         return xdst;
3002
3003 inc_error:
3004         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3005 error:
3006         xfrm_pols_put(pols, num_pols);
3007         return ERR_PTR(err);
3008 }
3009
3010 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3011                                         struct dst_entry *dst_orig)
3012 {
3013         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3014         struct dst_entry *ret;
3015
3016         if (!afinfo) {
3017                 dst_release(dst_orig);
3018                 return ERR_PTR(-EINVAL);
3019         } else {
3020                 ret = afinfo->blackhole_route(net, dst_orig);
3021         }
3022         rcu_read_unlock();
3023
3024         return ret;
3025 }
3026
3027 /* Finds/creates a bundle for given flow and if_id
3028  *
3029  * At the moment we eat a raw IP route. Mostly to speed up lookups
3030  * on interfaces with disabled IPsec.
3031  *
3032  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3033  * compatibility
3034  */
3035 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3036                                         struct dst_entry *dst_orig,
3037                                         const struct flowi *fl,
3038                                         const struct sock *sk,
3039                                         int flags, u32 if_id)
3040 {
3041         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3042         struct xfrm_dst *xdst;
3043         struct dst_entry *dst, *route;
3044         u16 family = dst_orig->ops->family;
3045         u8 dir = XFRM_POLICY_OUT;
3046         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3047
3048         dst = NULL;
3049         xdst = NULL;
3050         route = NULL;
3051
3052         sk = sk_const_to_full_sk(sk);
3053         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3054                 num_pols = 1;
3055                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3056                                                 if_id);
3057                 err = xfrm_expand_policies(fl, family, pols,
3058                                            &num_pols, &num_xfrms);
3059                 if (err < 0)
3060                         goto dropdst;
3061
3062                 if (num_pols) {
3063                         if (num_xfrms <= 0) {
3064                                 drop_pols = num_pols;
3065                                 goto no_transform;
3066                         }
3067
3068                         xdst = xfrm_resolve_and_create_bundle(
3069                                         pols, num_pols, fl,
3070                                         family, dst_orig);
3071
3072                         if (IS_ERR(xdst)) {
3073                                 xfrm_pols_put(pols, num_pols);
3074                                 err = PTR_ERR(xdst);
3075                                 if (err == -EREMOTE)
3076                                         goto nopol;
3077
3078                                 goto dropdst;
3079                         } else if (xdst == NULL) {
3080                                 num_xfrms = 0;
3081                                 drop_pols = num_pols;
3082                                 goto no_transform;
3083                         }
3084
3085                         route = xdst->route;
3086                 }
3087         }
3088
3089         if (xdst == NULL) {
3090                 struct xfrm_flo xflo;
3091
3092                 xflo.dst_orig = dst_orig;
3093                 xflo.flags = flags;
3094
3095                 /* To accelerate a bit...  */
3096                 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3097                                !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3098                         goto nopol;
3099
3100                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3101                 if (xdst == NULL)
3102                         goto nopol;
3103                 if (IS_ERR(xdst)) {
3104                         err = PTR_ERR(xdst);
3105                         goto dropdst;
3106                 }
3107
3108                 num_pols = xdst->num_pols;
3109                 num_xfrms = xdst->num_xfrms;
3110                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3111                 route = xdst->route;
3112         }
3113
3114         dst = &xdst->u.dst;
3115         if (route == NULL && num_xfrms > 0) {
3116                 /* The only case when xfrm_bundle_lookup() returns a
3117                  * bundle with null route, is when the template could
3118                  * not be resolved. It means policies are there, but
3119                  * bundle could not be created, since we don't yet
3120                  * have the xfrm_state's. We need to wait for KM to
3121                  * negotiate new SA's or bail out with error.*/
3122                 if (net->xfrm.sysctl_larval_drop) {
3123                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3124                         err = -EREMOTE;
3125                         goto error;
3126                 }
3127
3128                 err = -EAGAIN;
3129
3130                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3131                 goto error;
3132         }
3133
3134 no_transform:
3135         if (num_pols == 0)
3136                 goto nopol;
3137
3138         if ((flags & XFRM_LOOKUP_ICMP) &&
3139             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3140                 err = -ENOENT;
3141                 goto error;
3142         }
3143
3144         for (i = 0; i < num_pols; i++)
3145                 pols[i]->curlft.use_time = ktime_get_real_seconds();
3146
3147         if (num_xfrms < 0) {
3148                 /* Prohibit the flow */
3149                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3150                 err = -EPERM;
3151                 goto error;
3152         } else if (num_xfrms > 0) {
3153                 /* Flow transformed */
3154                 dst_release(dst_orig);
3155         } else {
3156                 /* Flow passes untransformed */
3157                 dst_release(dst);
3158                 dst = dst_orig;
3159         }
3160 ok:
3161         xfrm_pols_put(pols, drop_pols);
3162         if (dst && dst->xfrm &&
3163             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3164                 dst->flags |= DST_XFRM_TUNNEL;
3165         return dst;
3166
3167 nopol:
3168         if (!(flags & XFRM_LOOKUP_ICMP)) {
3169                 dst = dst_orig;
3170                 goto ok;
3171         }
3172         err = -ENOENT;
3173 error:
3174         dst_release(dst);
3175 dropdst:
3176         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3177                 dst_release(dst_orig);
3178         xfrm_pols_put(pols, drop_pols);
3179         return ERR_PTR(err);
3180 }
3181 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3182
3183 /* Main function: finds/creates a bundle for given flow.
3184  *
3185  * At the moment we eat a raw IP route. Mostly to speed up lookups
3186  * on interfaces with disabled IPsec.
3187  */
3188 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3189                               const struct flowi *fl, const struct sock *sk,
3190                               int flags)
3191 {
3192         return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3193 }
3194 EXPORT_SYMBOL(xfrm_lookup);
3195
3196 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3197  * Otherwise we may send out blackholed packets.
3198  */
3199 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3200                                     const struct flowi *fl,
3201                                     const struct sock *sk, int flags)
3202 {
3203         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3204                                             flags | XFRM_LOOKUP_QUEUE |
3205                                             XFRM_LOOKUP_KEEP_DST_REF);
3206
3207         if (PTR_ERR(dst) == -EREMOTE)
3208                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3209
3210         if (IS_ERR(dst))
3211                 dst_release(dst_orig);
3212
3213         return dst;
3214 }
3215 EXPORT_SYMBOL(xfrm_lookup_route);
3216
3217 static inline int
3218 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3219 {
3220         struct sec_path *sp = skb_sec_path(skb);
3221         struct xfrm_state *x;
3222
3223         if (!sp || idx < 0 || idx >= sp->len)
3224                 return 0;
3225         x = sp->xvec[idx];
3226         if (!x->type->reject)
3227                 return 0;
3228         return x->type->reject(x, skb, fl);
3229 }
3230
3231 /* When skb is transformed back to its "native" form, we have to
3232  * check policy restrictions. At the moment we make this in maximally
3233  * stupid way. Shame on me. :-) Of course, connected sockets must
3234  * have policy cached at them.
3235  */
3236
3237 static inline int
3238 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3239               unsigned short family)
3240 {
3241         if (xfrm_state_kern(x))
3242                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3243         return  x->id.proto == tmpl->id.proto &&
3244                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3245                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3246                 x->props.mode == tmpl->mode &&
3247                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3248                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3249                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3250                   xfrm_state_addr_cmp(tmpl, x, family));
3251 }
3252
3253 /*
3254  * 0 or more than 0 is returned when validation is succeeded (either bypass
3255  * because of optional transport mode, or next index of the matched secpath
3256  * state with the template.
3257  * -1 is returned when no matching template is found.
3258  * Otherwise "-2 - errored_index" is returned.
3259  */
3260 static inline int
3261 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3262                unsigned short family)
3263 {
3264         int idx = start;
3265
3266         if (tmpl->optional) {
3267                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3268                         return start;
3269         } else
3270                 start = -1;
3271         for (; idx < sp->len; idx++) {
3272                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3273                         return ++idx;
3274                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3275                         if (start == -1)
3276                                 start = -2-idx;
3277                         break;
3278                 }
3279         }
3280         return start;
3281 }
3282
3283 static void
3284 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3285 {
3286         const struct iphdr *iph = ip_hdr(skb);
3287         int ihl = iph->ihl;
3288         u8 *xprth = skb_network_header(skb) + ihl * 4;
3289         struct flowi4 *fl4 = &fl->u.ip4;
3290         int oif = 0;
3291
3292         if (skb_dst(skb) && skb_dst(skb)->dev)
3293                 oif = skb_dst(skb)->dev->ifindex;
3294
3295         memset(fl4, 0, sizeof(struct flowi4));
3296         fl4->flowi4_mark = skb->mark;
3297         fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3298
3299         fl4->flowi4_proto = iph->protocol;
3300         fl4->daddr = reverse ? iph->saddr : iph->daddr;
3301         fl4->saddr = reverse ? iph->daddr : iph->saddr;
3302         fl4->flowi4_tos = iph->tos;
3303
3304         if (!ip_is_fragment(iph)) {
3305                 switch (iph->protocol) {
3306                 case IPPROTO_UDP:
3307                 case IPPROTO_UDPLITE:
3308                 case IPPROTO_TCP:
3309                 case IPPROTO_SCTP:
3310                 case IPPROTO_DCCP:
3311                         if (xprth + 4 < skb->data ||
3312                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
3313                                 __be16 *ports;
3314
3315                                 xprth = skb_network_header(skb) + ihl * 4;
3316                                 ports = (__be16 *)xprth;
3317
3318                                 fl4->fl4_sport = ports[!!reverse];
3319                                 fl4->fl4_dport = ports[!reverse];
3320                         }
3321                         break;
3322                 case IPPROTO_ICMP:
3323                         if (xprth + 2 < skb->data ||
3324                             pskb_may_pull(skb, xprth + 2 - skb->data)) {
3325                                 u8 *icmp;
3326
3327                                 xprth = skb_network_header(skb) + ihl * 4;
3328                                 icmp = xprth;
3329
3330                                 fl4->fl4_icmp_type = icmp[0];
3331                                 fl4->fl4_icmp_code = icmp[1];
3332                         }
3333                         break;
3334                 case IPPROTO_GRE:
3335                         if (xprth + 12 < skb->data ||
3336                             pskb_may_pull(skb, xprth + 12 - skb->data)) {
3337                                 __be16 *greflags;
3338                                 __be32 *gre_hdr;
3339
3340                                 xprth = skb_network_header(skb) + ihl * 4;
3341                                 greflags = (__be16 *)xprth;
3342                                 gre_hdr = (__be32 *)xprth;
3343
3344                                 if (greflags[0] & GRE_KEY) {
3345                                         if (greflags[0] & GRE_CSUM)
3346                                                 gre_hdr++;
3347                                         fl4->fl4_gre_key = gre_hdr[1];
3348                                 }
3349                         }
3350                         break;
3351                 default:
3352                         break;
3353                 }
3354         }
3355 }
3356
3357 #if IS_ENABLED(CONFIG_IPV6)
3358 static void
3359 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3360 {
3361         struct flowi6 *fl6 = &fl->u.ip6;
3362         int onlyproto = 0;
3363         const struct ipv6hdr *hdr = ipv6_hdr(skb);
3364         u32 offset = sizeof(*hdr);
3365         struct ipv6_opt_hdr *exthdr;
3366         const unsigned char *nh = skb_network_header(skb);
3367         u16 nhoff = IP6CB(skb)->nhoff;
3368         int oif = 0;
3369         u8 nexthdr;
3370
3371         if (!nhoff)
3372                 nhoff = offsetof(struct ipv6hdr, nexthdr);
3373
3374         nexthdr = nh[nhoff];
3375
3376         if (skb_dst(skb) && skb_dst(skb)->dev)
3377                 oif = skb_dst(skb)->dev->ifindex;
3378
3379         memset(fl6, 0, sizeof(struct flowi6));
3380         fl6->flowi6_mark = skb->mark;
3381         fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3382
3383         fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3384         fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3385
3386         while (nh + offset + sizeof(*exthdr) < skb->data ||
3387                pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3388                 nh = skb_network_header(skb);
3389                 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3390
3391                 switch (nexthdr) {
3392                 case NEXTHDR_FRAGMENT:
3393                         onlyproto = 1;
3394                         fallthrough;
3395                 case NEXTHDR_ROUTING:
3396                 case NEXTHDR_HOP:
3397                 case NEXTHDR_DEST:
3398                         offset += ipv6_optlen(exthdr);
3399                         nexthdr = exthdr->nexthdr;
3400                         exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3401                         break;
3402                 case IPPROTO_UDP:
3403                 case IPPROTO_UDPLITE:
3404                 case IPPROTO_TCP:
3405                 case IPPROTO_SCTP:
3406                 case IPPROTO_DCCP:
3407                         if (!onlyproto && (nh + offset + 4 < skb->data ||
3408                              pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3409                                 __be16 *ports;
3410
3411                                 nh = skb_network_header(skb);
3412                                 ports = (__be16 *)(nh + offset);
3413                                 fl6->fl6_sport = ports[!!reverse];
3414                                 fl6->fl6_dport = ports[!reverse];
3415                         }
3416                         fl6->flowi6_proto = nexthdr;
3417                         return;
3418                 case IPPROTO_ICMPV6:
3419                         if (!onlyproto && (nh + offset + 2 < skb->data ||
3420                             pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3421                                 u8 *icmp;
3422
3423                                 nh = skb_network_header(skb);
3424                                 icmp = (u8 *)(nh + offset);
3425                                 fl6->fl6_icmp_type = icmp[0];
3426                                 fl6->fl6_icmp_code = icmp[1];
3427                         }
3428                         fl6->flowi6_proto = nexthdr;
3429                         return;
3430 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3431                 case IPPROTO_MH:
3432                         offset += ipv6_optlen(exthdr);
3433                         if (!onlyproto && (nh + offset + 3 < skb->data ||
3434                             pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3435                                 struct ip6_mh *mh;
3436
3437                                 nh = skb_network_header(skb);
3438                                 mh = (struct ip6_mh *)(nh + offset);
3439                                 fl6->fl6_mh_type = mh->ip6mh_type;
3440                         }
3441                         fl6->flowi6_proto = nexthdr;
3442                         return;
3443 #endif
3444                 default:
3445                         fl6->flowi6_proto = nexthdr;
3446                         return;
3447                 }
3448         }
3449 }
3450 #endif
3451
3452 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3453                           unsigned int family, int reverse)
3454 {
3455         switch (family) {
3456         case AF_INET:
3457                 decode_session4(skb, fl, reverse);
3458                 break;
3459 #if IS_ENABLED(CONFIG_IPV6)
3460         case AF_INET6:
3461                 decode_session6(skb, fl, reverse);
3462                 break;
3463 #endif
3464         default:
3465                 return -EAFNOSUPPORT;
3466         }
3467
3468         return security_xfrm_decode_session(skb, &fl->flowi_secid);
3469 }
3470 EXPORT_SYMBOL(__xfrm_decode_session);
3471
3472 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3473 {
3474         for (; k < sp->len; k++) {
3475                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3476                         *idxp = k;
3477                         return 1;
3478                 }
3479         }
3480
3481         return 0;
3482 }
3483
3484 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3485                         unsigned short family)
3486 {
3487         struct net *net = dev_net(skb->dev);
3488         struct xfrm_policy *pol;
3489         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3490         int npols = 0;
3491         int xfrm_nr;
3492         int pi;
3493         int reverse;
3494         struct flowi fl;
3495         int xerr_idx = -1;
3496         const struct xfrm_if_cb *ifcb;
3497         struct sec_path *sp;
3498         struct xfrm_if *xi;
3499         u32 if_id = 0;
3500
3501         rcu_read_lock();
3502         ifcb = xfrm_if_get_cb();
3503
3504         if (ifcb) {
3505                 xi = ifcb->decode_session(skb, family);
3506                 if (xi) {
3507                         if_id = xi->p.if_id;
3508                         net = xi->net;
3509                 }
3510         }
3511         rcu_read_unlock();
3512
3513         reverse = dir & ~XFRM_POLICY_MASK;
3514         dir &= XFRM_POLICY_MASK;
3515
3516         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3517                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3518                 return 0;
3519         }
3520
3521         nf_nat_decode_session(skb, &fl, family);
3522
3523         /* First, check used SA against their selectors. */
3524         sp = skb_sec_path(skb);
3525         if (sp) {
3526                 int i;
3527
3528                 for (i = sp->len - 1; i >= 0; i--) {
3529                         struct xfrm_state *x = sp->xvec[i];
3530                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
3531                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3532                                 return 0;
3533                         }
3534                 }
3535         }
3536
3537         pol = NULL;
3538         sk = sk_to_full_sk(sk);
3539         if (sk && sk->sk_policy[dir]) {
3540                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3541                 if (IS_ERR(pol)) {
3542                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3543                         return 0;
3544                 }
3545         }
3546
3547         if (!pol)
3548                 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3549
3550         if (IS_ERR(pol)) {
3551                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3552                 return 0;
3553         }
3554
3555         if (!pol) {
3556                 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3557                         xfrm_secpath_reject(xerr_idx, skb, &fl);
3558                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3559                         return 0;
3560                 }
3561                 return 1;
3562         }
3563
3564         pol->curlft.use_time = ktime_get_real_seconds();
3565
3566         pols[0] = pol;
3567         npols++;
3568 #ifdef CONFIG_XFRM_SUB_POLICY
3569         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3570                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3571                                                     &fl, family,
3572                                                     XFRM_POLICY_IN, if_id);
3573                 if (pols[1]) {
3574                         if (IS_ERR(pols[1])) {
3575                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3576                                 return 0;
3577                         }
3578                         pols[1]->curlft.use_time = ktime_get_real_seconds();
3579                         npols++;
3580                 }
3581         }
3582 #endif
3583
3584         if (pol->action == XFRM_POLICY_ALLOW) {
3585                 static struct sec_path dummy;
3586                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3587                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3588                 struct xfrm_tmpl **tpp = tp;
3589                 int ti = 0;
3590                 int i, k;
3591
3592                 sp = skb_sec_path(skb);
3593                 if (!sp)
3594                         sp = &dummy;
3595
3596                 for (pi = 0; pi < npols; pi++) {
3597                         if (pols[pi] != pol &&
3598                             pols[pi]->action != XFRM_POLICY_ALLOW) {
3599                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3600                                 goto reject;
3601                         }
3602                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3603                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3604                                 goto reject_error;
3605                         }
3606                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
3607                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3608                 }
3609                 xfrm_nr = ti;
3610                 if (npols > 1) {
3611                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3612                         tpp = stp;
3613                 }
3614
3615                 /* For each tunnel xfrm, find the first matching tmpl.
3616                  * For each tmpl before that, find corresponding xfrm.
3617                  * Order is _important_. Later we will implement
3618                  * some barriers, but at the moment barriers
3619                  * are implied between each two transformations.
3620                  */
3621                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3622                         k = xfrm_policy_ok(tpp[i], sp, k, family);
3623                         if (k < 0) {
3624                                 if (k < -1)
3625                                         /* "-2 - errored_index" returned */
3626                                         xerr_idx = -(2+k);
3627                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3628                                 goto reject;
3629                         }
3630                 }
3631
3632                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3633                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3634                         goto reject;
3635                 }
3636
3637                 xfrm_pols_put(pols, npols);
3638                 return 1;
3639         }
3640         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3641
3642 reject:
3643         xfrm_secpath_reject(xerr_idx, skb, &fl);
3644 reject_error:
3645         xfrm_pols_put(pols, npols);
3646         return 0;
3647 }
3648 EXPORT_SYMBOL(__xfrm_policy_check);
3649
3650 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3651 {
3652         struct net *net = dev_net(skb->dev);
3653         struct flowi fl;
3654         struct dst_entry *dst;
3655         int res = 1;
3656
3657         if (xfrm_decode_session(skb, &fl, family) < 0) {
3658                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3659                 return 0;
3660         }
3661
3662         skb_dst_force(skb);
3663         if (!skb_dst(skb)) {
3664                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3665                 return 0;
3666         }
3667
3668         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3669         if (IS_ERR(dst)) {
3670                 res = 0;
3671                 dst = NULL;
3672         }
3673         skb_dst_set(skb, dst);
3674         return res;
3675 }
3676 EXPORT_SYMBOL(__xfrm_route_forward);
3677
3678 /* Optimize later using cookies and generation ids. */
3679
3680 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3681 {
3682         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3683          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3684          * get validated by dst_ops->check on every use.  We do this
3685          * because when a normal route referenced by an XFRM dst is
3686          * obsoleted we do not go looking around for all parent
3687          * referencing XFRM dsts so that we can invalidate them.  It
3688          * is just too much work.  Instead we make the checks here on
3689          * every use.  For example:
3690          *
3691          *      XFRM dst A --> IPv4 dst X
3692          *
3693          * X is the "xdst->route" of A (X is also the "dst->path" of A
3694          * in this example).  If X is marked obsolete, "A" will not
3695          * notice.  That's what we are validating here via the
3696          * stale_bundle() check.
3697          *
3698          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3699          * be marked on it.
3700          * This will force stale_bundle() to fail on any xdst bundle with
3701          * this dst linked in it.
3702          */
3703         if (dst->obsolete < 0 && !stale_bundle(dst))
3704                 return dst;
3705
3706         return NULL;
3707 }
3708
3709 static int stale_bundle(struct dst_entry *dst)
3710 {
3711         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3712 }
3713
3714 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3715 {
3716         while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3717                 dst->dev = dev_net(dev)->loopback_dev;
3718                 dev_hold(dst->dev);
3719                 dev_put(dev);
3720         }
3721 }
3722 EXPORT_SYMBOL(xfrm_dst_ifdown);
3723
3724 static void xfrm_link_failure(struct sk_buff *skb)
3725 {
3726         /* Impossible. Such dst must be popped before reaches point of failure. */
3727 }
3728
3729 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3730 {
3731         if (dst) {
3732                 if (dst->obsolete) {
3733                         dst_release(dst);
3734                         dst = NULL;
3735                 }
3736         }
3737         return dst;
3738 }
3739
3740 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3741 {
3742         while (nr--) {
3743                 struct xfrm_dst *xdst = bundle[nr];
3744                 u32 pmtu, route_mtu_cached;
3745                 struct dst_entry *dst;
3746
3747                 dst = &xdst->u.dst;
3748                 pmtu = dst_mtu(xfrm_dst_child(dst));
3749                 xdst->child_mtu_cached = pmtu;
3750
3751                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3752
3753                 route_mtu_cached = dst_mtu(xdst->route);
3754                 xdst->route_mtu_cached = route_mtu_cached;
3755
3756                 if (pmtu > route_mtu_cached)
3757                         pmtu = route_mtu_cached;
3758
3759                 dst_metric_set(dst, RTAX_MTU, pmtu);
3760         }
3761 }
3762
3763 /* Check that the bundle accepts the flow and its components are
3764  * still valid.
3765  */
3766
3767 static int xfrm_bundle_ok(struct xfrm_dst *first)
3768 {
3769         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3770         struct dst_entry *dst = &first->u.dst;
3771         struct xfrm_dst *xdst;
3772         int start_from, nr;
3773         u32 mtu;
3774
3775         if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3776             (dst->dev && !netif_running(dst->dev)))
3777                 return 0;
3778
3779         if (dst->flags & DST_XFRM_QUEUE)
3780                 return 1;
3781
3782         start_from = nr = 0;
3783         do {
3784                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3785
3786                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3787                         return 0;
3788                 if (xdst->xfrm_genid != dst->xfrm->genid)
3789                         return 0;
3790                 if (xdst->num_pols > 0 &&
3791                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3792                         return 0;
3793
3794                 bundle[nr++] = xdst;
3795
3796                 mtu = dst_mtu(xfrm_dst_child(dst));
3797                 if (xdst->child_mtu_cached != mtu) {
3798                         start_from = nr;
3799                         xdst->child_mtu_cached = mtu;
3800                 }
3801
3802                 if (!dst_check(xdst->route, xdst->route_cookie))
3803                         return 0;
3804                 mtu = dst_mtu(xdst->route);
3805                 if (xdst->route_mtu_cached != mtu) {
3806                         start_from = nr;
3807                         xdst->route_mtu_cached = mtu;
3808                 }
3809
3810                 dst = xfrm_dst_child(dst);
3811         } while (dst->xfrm);
3812
3813         if (likely(!start_from))
3814                 return 1;
3815
3816         xdst = bundle[start_from - 1];
3817         mtu = xdst->child_mtu_cached;
3818         while (start_from--) {
3819                 dst = &xdst->u.dst;
3820
3821                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3822                 if (mtu > xdst->route_mtu_cached)
3823                         mtu = xdst->route_mtu_cached;
3824                 dst_metric_set(dst, RTAX_MTU, mtu);
3825                 if (!start_from)
3826                         break;
3827
3828                 xdst = bundle[start_from - 1];
3829                 xdst->child_mtu_cached = mtu;
3830         }
3831
3832         return 1;
3833 }
3834
3835 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3836 {
3837         return dst_metric_advmss(xfrm_dst_path(dst));
3838 }
3839
3840 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3841 {
3842         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3843
3844         return mtu ? : dst_mtu(xfrm_dst_path(dst));
3845 }
3846
3847 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3848                                         const void *daddr)
3849 {
3850         while (dst->xfrm) {
3851                 const struct xfrm_state *xfrm = dst->xfrm;
3852
3853                 dst = xfrm_dst_child(dst);
3854
3855                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3856                         continue;
3857                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3858                         daddr = xfrm->coaddr;
3859                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3860                         daddr = &xfrm->id.daddr;
3861         }
3862         return daddr;
3863 }
3864
3865 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3866                                            struct sk_buff *skb,
3867                                            const void *daddr)
3868 {
3869         const struct dst_entry *path = xfrm_dst_path(dst);
3870
3871         if (!skb)
3872                 daddr = xfrm_get_dst_nexthop(dst, daddr);
3873         return path->ops->neigh_lookup(path, skb, daddr);
3874 }
3875
3876 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3877 {
3878         const struct dst_entry *path = xfrm_dst_path(dst);
3879
3880         daddr = xfrm_get_dst_nexthop(dst, daddr);
3881         path->ops->confirm_neigh(path, daddr);
3882 }
3883
3884 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3885 {
3886         int err = 0;
3887
3888         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3889                 return -EAFNOSUPPORT;
3890
3891         spin_lock(&xfrm_policy_afinfo_lock);
3892         if (unlikely(xfrm_policy_afinfo[family] != NULL))
3893                 err = -EEXIST;
3894         else {
3895                 struct dst_ops *dst_ops = afinfo->dst_ops;
3896                 if (likely(dst_ops->kmem_cachep == NULL))
3897                         dst_ops->kmem_cachep = xfrm_dst_cache;
3898                 if (likely(dst_ops->check == NULL))
3899                         dst_ops->check = xfrm_dst_check;
3900                 if (likely(dst_ops->default_advmss == NULL))
3901                         dst_ops->default_advmss = xfrm_default_advmss;
3902                 if (likely(dst_ops->mtu == NULL))
3903                         dst_ops->mtu = xfrm_mtu;
3904                 if (likely(dst_ops->negative_advice == NULL))
3905                         dst_ops->negative_advice = xfrm_negative_advice;
3906                 if (likely(dst_ops->link_failure == NULL))
3907                         dst_ops->link_failure = xfrm_link_failure;
3908                 if (likely(dst_ops->neigh_lookup == NULL))
3909                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
3910                 if (likely(!dst_ops->confirm_neigh))
3911                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
3912                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3913         }
3914         spin_unlock(&xfrm_policy_afinfo_lock);
3915
3916         return err;
3917 }
3918 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3919
3920 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3921 {
3922         struct dst_ops *dst_ops = afinfo->dst_ops;
3923         int i;
3924
3925         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3926                 if (xfrm_policy_afinfo[i] != afinfo)
3927                         continue;
3928                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3929                 break;
3930         }
3931
3932         synchronize_rcu();
3933
3934         dst_ops->kmem_cachep = NULL;
3935         dst_ops->check = NULL;
3936         dst_ops->negative_advice = NULL;
3937         dst_ops->link_failure = NULL;
3938 }
3939 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3940
3941 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3942 {
3943         spin_lock(&xfrm_if_cb_lock);
3944         rcu_assign_pointer(xfrm_if_cb, ifcb);
3945         spin_unlock(&xfrm_if_cb_lock);
3946 }
3947 EXPORT_SYMBOL(xfrm_if_register_cb);
3948
3949 void xfrm_if_unregister_cb(void)
3950 {
3951         RCU_INIT_POINTER(xfrm_if_cb, NULL);
3952         synchronize_rcu();
3953 }
3954 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3955
3956 #ifdef CONFIG_XFRM_STATISTICS
3957 static int __net_init xfrm_statistics_init(struct net *net)
3958 {
3959         int rv;
3960         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3961         if (!net->mib.xfrm_statistics)
3962                 return -ENOMEM;
3963         rv = xfrm_proc_init(net);
3964         if (rv < 0)
3965                 free_percpu(net->mib.xfrm_statistics);
3966         return rv;
3967 }
3968
3969 static void xfrm_statistics_fini(struct net *net)
3970 {
3971         xfrm_proc_fini(net);
3972         free_percpu(net->mib.xfrm_statistics);
3973 }
3974 #else
3975 static int __net_init xfrm_statistics_init(struct net *net)
3976 {
3977         return 0;
3978 }
3979
3980 static void xfrm_statistics_fini(struct net *net)
3981 {
3982 }
3983 #endif
3984
3985 static int __net_init xfrm_policy_init(struct net *net)
3986 {
3987         unsigned int hmask, sz;
3988         int dir, err;
3989
3990         if (net_eq(net, &init_net)) {
3991                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
3992                                            sizeof(struct xfrm_dst),
3993                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3994                                            NULL);
3995                 err = rhashtable_init(&xfrm_policy_inexact_table,
3996                                       &xfrm_pol_inexact_params);
3997                 BUG_ON(err);
3998         }
3999
4000         hmask = 8 - 1;
4001         sz = (hmask+1) * sizeof(struct hlist_head);
4002
4003         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4004         if (!net->xfrm.policy_byidx)
4005                 goto out_byidx;
4006         net->xfrm.policy_idx_hmask = hmask;
4007
4008         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4009                 struct xfrm_policy_hash *htab;
4010
4011                 net->xfrm.policy_count[dir] = 0;
4012                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4013                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4014
4015                 htab = &net->xfrm.policy_bydst[dir];
4016                 htab->table = xfrm_hash_alloc(sz);
4017                 if (!htab->table)
4018                         goto out_bydst;
4019                 htab->hmask = hmask;
4020                 htab->dbits4 = 32;
4021                 htab->sbits4 = 32;
4022                 htab->dbits6 = 128;
4023                 htab->sbits6 = 128;
4024         }
4025         net->xfrm.policy_hthresh.lbits4 = 32;
4026         net->xfrm.policy_hthresh.rbits4 = 32;
4027         net->xfrm.policy_hthresh.lbits6 = 128;
4028         net->xfrm.policy_hthresh.rbits6 = 128;
4029
4030         seqlock_init(&net->xfrm.policy_hthresh.lock);
4031
4032         INIT_LIST_HEAD(&net->xfrm.policy_all);
4033         INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4034         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4035         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4036         return 0;
4037
4038 out_bydst:
4039         for (dir--; dir >= 0; dir--) {
4040                 struct xfrm_policy_hash *htab;
4041
4042                 htab = &net->xfrm.policy_bydst[dir];
4043                 xfrm_hash_free(htab->table, sz);
4044         }
4045         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4046 out_byidx:
4047         return -ENOMEM;
4048 }
4049
4050 static void xfrm_policy_fini(struct net *net)
4051 {
4052         struct xfrm_pol_inexact_bin *b, *t;
4053         unsigned int sz;
4054         int dir;
4055
4056         flush_work(&net->xfrm.policy_hash_work);
4057 #ifdef CONFIG_XFRM_SUB_POLICY
4058         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4059 #endif
4060         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4061
4062         WARN_ON(!list_empty(&net->xfrm.policy_all));
4063
4064         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4065                 struct xfrm_policy_hash *htab;
4066
4067                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4068
4069                 htab = &net->xfrm.policy_bydst[dir];
4070                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4071                 WARN_ON(!hlist_empty(htab->table));
4072                 xfrm_hash_free(htab->table, sz);
4073         }
4074
4075         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4076         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4077         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4078
4079         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4080         list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4081                 __xfrm_policy_inexact_prune_bin(b, true);
4082         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4083 }
4084
4085 static int __net_init xfrm_net_init(struct net *net)
4086 {
4087         int rv;
4088
4089         /* Initialize the per-net locks here */
4090         spin_lock_init(&net->xfrm.xfrm_state_lock);
4091         spin_lock_init(&net->xfrm.xfrm_policy_lock);
4092         mutex_init(&net->xfrm.xfrm_cfg_mutex);
4093
4094         rv = xfrm_statistics_init(net);
4095         if (rv < 0)
4096                 goto out_statistics;
4097         rv = xfrm_state_init(net);
4098         if (rv < 0)
4099                 goto out_state;
4100         rv = xfrm_policy_init(net);
4101         if (rv < 0)
4102                 goto out_policy;
4103         rv = xfrm_sysctl_init(net);
4104         if (rv < 0)
4105                 goto out_sysctl;
4106
4107         return 0;
4108
4109 out_sysctl:
4110         xfrm_policy_fini(net);
4111 out_policy:
4112         xfrm_state_fini(net);
4113 out_state:
4114         xfrm_statistics_fini(net);
4115 out_statistics:
4116         return rv;
4117 }
4118
4119 static void __net_exit xfrm_net_exit(struct net *net)
4120 {
4121         xfrm_sysctl_fini(net);
4122         xfrm_policy_fini(net);
4123         xfrm_state_fini(net);
4124         xfrm_statistics_fini(net);
4125 }
4126
4127 static struct pernet_operations __net_initdata xfrm_net_ops = {
4128         .init = xfrm_net_init,
4129         .exit = xfrm_net_exit,
4130 };
4131
4132 void __init xfrm_init(void)
4133 {
4134         register_pernet_subsys(&xfrm_net_ops);
4135         xfrm_dev_init();
4136         seqcount_mutex_init(&xfrm_policy_hash_generation, &hash_resize_mutex);
4137         xfrm_input_init();
4138
4139 #ifdef CONFIG_XFRM_ESPINTCP
4140         espintcp_init();
4141 #endif
4142 }
4143
4144 #ifdef CONFIG_AUDITSYSCALL
4145 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4146                                          struct audit_buffer *audit_buf)
4147 {
4148         struct xfrm_sec_ctx *ctx = xp->security;
4149         struct xfrm_selector *sel = &xp->selector;
4150
4151         if (ctx)
4152                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4153                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4154
4155         switch (sel->family) {
4156         case AF_INET:
4157                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4158                 if (sel->prefixlen_s != 32)
4159                         audit_log_format(audit_buf, " src_prefixlen=%d",
4160                                          sel->prefixlen_s);
4161                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4162                 if (sel->prefixlen_d != 32)
4163                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4164                                          sel->prefixlen_d);
4165                 break;
4166         case AF_INET6:
4167                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4168                 if (sel->prefixlen_s != 128)
4169                         audit_log_format(audit_buf, " src_prefixlen=%d",
4170                                          sel->prefixlen_s);
4171                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4172                 if (sel->prefixlen_d != 128)
4173                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4174                                          sel->prefixlen_d);
4175                 break;
4176         }
4177 }
4178
4179 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4180 {
4181         struct audit_buffer *audit_buf;
4182
4183         audit_buf = xfrm_audit_start("SPD-add");
4184         if (audit_buf == NULL)
4185                 return;
4186         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4187         audit_log_format(audit_buf, " res=%u", result);
4188         xfrm_audit_common_policyinfo(xp, audit_buf);
4189         audit_log_end(audit_buf);
4190 }
4191 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4192
4193 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4194                               bool task_valid)
4195 {
4196         struct audit_buffer *audit_buf;
4197
4198         audit_buf = xfrm_audit_start("SPD-delete");
4199         if (audit_buf == NULL)
4200                 return;
4201         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4202         audit_log_format(audit_buf, " res=%u", result);
4203         xfrm_audit_common_policyinfo(xp, audit_buf);
4204         audit_log_end(audit_buf);
4205 }
4206 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4207 #endif
4208
4209 #ifdef CONFIG_XFRM_MIGRATE
4210 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4211                                         const struct xfrm_selector *sel_tgt)
4212 {
4213         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4214                 if (sel_tgt->family == sel_cmp->family &&
4215                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4216                                     sel_cmp->family) &&
4217                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4218                                     sel_cmp->family) &&
4219                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4220                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4221                         return true;
4222                 }
4223         } else {
4224                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4225                         return true;
4226                 }
4227         }
4228         return false;
4229 }
4230
4231 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4232                                                     u8 dir, u8 type, struct net *net)
4233 {
4234         struct xfrm_policy *pol, *ret = NULL;
4235         struct hlist_head *chain;
4236         u32 priority = ~0U;
4237
4238         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4239         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4240         hlist_for_each_entry(pol, chain, bydst) {
4241                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4242                     pol->type == type) {
4243                         ret = pol;
4244                         priority = ret->priority;
4245                         break;
4246                 }
4247         }
4248         chain = &net->xfrm.policy_inexact[dir];
4249         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4250                 if ((pol->priority >= priority) && ret)
4251                         break;
4252
4253                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4254                     pol->type == type) {
4255                         ret = pol;
4256                         break;
4257                 }
4258         }
4259
4260         xfrm_pol_hold(ret);
4261
4262         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4263
4264         return ret;
4265 }
4266
4267 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4268 {
4269         int match = 0;
4270
4271         if (t->mode == m->mode && t->id.proto == m->proto &&
4272             (m->reqid == 0 || t->reqid == m->reqid)) {
4273                 switch (t->mode) {
4274                 case XFRM_MODE_TUNNEL:
4275                 case XFRM_MODE_BEET:
4276                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4277                                             m->old_family) &&
4278                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
4279                                             m->old_family)) {
4280                                 match = 1;
4281                         }
4282                         break;
4283                 case XFRM_MODE_TRANSPORT:
4284                         /* in case of transport mode, template does not store
4285                            any IP addresses, hence we just compare mode and
4286                            protocol */
4287                         match = 1;
4288                         break;
4289                 default:
4290                         break;
4291                 }
4292         }
4293         return match;
4294 }
4295
4296 /* update endpoint address(es) of template(s) */
4297 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4298                                struct xfrm_migrate *m, int num_migrate)
4299 {
4300         struct xfrm_migrate *mp;
4301         int i, j, n = 0;
4302
4303         write_lock_bh(&pol->lock);
4304         if (unlikely(pol->walk.dead)) {
4305                 /* target policy has been deleted */
4306                 write_unlock_bh(&pol->lock);
4307                 return -ENOENT;
4308         }
4309
4310         for (i = 0; i < pol->xfrm_nr; i++) {
4311                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4312                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4313                                 continue;
4314                         n++;
4315                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4316                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4317                                 continue;
4318                         /* update endpoints */
4319                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4320                                sizeof(pol->xfrm_vec[i].id.daddr));
4321                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4322                                sizeof(pol->xfrm_vec[i].saddr));
4323                         pol->xfrm_vec[i].encap_family = mp->new_family;
4324                         /* flush bundles */
4325                         atomic_inc(&pol->genid);
4326                 }
4327         }
4328
4329         write_unlock_bh(&pol->lock);
4330
4331         if (!n)
4332                 return -ENODATA;
4333
4334         return 0;
4335 }
4336
4337 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4338 {
4339         int i, j;
4340
4341         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4342                 return -EINVAL;
4343
4344         for (i = 0; i < num_migrate; i++) {
4345                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4346                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4347                         return -EINVAL;
4348
4349                 /* check if there is any duplicated entry */
4350                 for (j = i + 1; j < num_migrate; j++) {
4351                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4352                                     sizeof(m[i].old_daddr)) &&
4353                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4354                                     sizeof(m[i].old_saddr)) &&
4355                             m[i].proto == m[j].proto &&
4356                             m[i].mode == m[j].mode &&
4357                             m[i].reqid == m[j].reqid &&
4358                             m[i].old_family == m[j].old_family)
4359                                 return -EINVAL;
4360                 }
4361         }
4362
4363         return 0;
4364 }
4365
4366 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4367                  struct xfrm_migrate *m, int num_migrate,
4368                  struct xfrm_kmaddress *k, struct net *net,
4369                  struct xfrm_encap_tmpl *encap)
4370 {
4371         int i, err, nx_cur = 0, nx_new = 0;
4372         struct xfrm_policy *pol = NULL;
4373         struct xfrm_state *x, *xc;
4374         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4375         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4376         struct xfrm_migrate *mp;
4377
4378         /* Stage 0 - sanity checks */
4379         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4380                 goto out;
4381
4382         if (dir >= XFRM_POLICY_MAX) {
4383                 err = -EINVAL;
4384                 goto out;
4385         }
4386
4387         /* Stage 1 - find policy */
4388         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4389                 err = -ENOENT;
4390                 goto out;
4391         }
4392
4393         /* Stage 2 - find and update state(s) */
4394         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4395                 if ((x = xfrm_migrate_state_find(mp, net))) {
4396                         x_cur[nx_cur] = x;
4397                         nx_cur++;
4398                         xc = xfrm_state_migrate(x, mp, encap);
4399                         if (xc) {
4400                                 x_new[nx_new] = xc;
4401                                 nx_new++;
4402                         } else {
4403                                 err = -ENODATA;
4404                                 goto restore_state;
4405                         }
4406                 }
4407         }
4408
4409         /* Stage 3 - update policy */
4410         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4411                 goto restore_state;
4412
4413         /* Stage 4 - delete old state(s) */
4414         if (nx_cur) {
4415                 xfrm_states_put(x_cur, nx_cur);
4416                 xfrm_states_delete(x_cur, nx_cur);
4417         }
4418
4419         /* Stage 5 - announce */
4420         km_migrate(sel, dir, type, m, num_migrate, k, encap);
4421
4422         xfrm_pol_put(pol);
4423
4424         return 0;
4425 out:
4426         return err;
4427
4428 restore_state:
4429         if (pol)
4430                 xfrm_pol_put(pol);
4431         if (nx_cur)
4432                 xfrm_states_put(x_cur, nx_cur);
4433         if (nx_new)
4434                 xfrm_states_delete(x_new, nx_new);
4435
4436         return err;
4437 }
4438 EXPORT_SYMBOL(xfrm_migrate);
4439 #endif