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
 159 static struct rhashtable xfrm_policy_inexact_table;
 160 static const struct rhashtable_params xfrm_pol_inexact_params;
 161
 162 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
 163 static int stale_bundle(struct dst_entry *dst);
 164 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
 165 static void xfrm_policy_queue_process(struct timer_list *t);
 166
 167 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
 168 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
 169                                                 int dir);
 170
 171 static struct xfrm_pol_inexact_bin *
 172 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
 173                            u32 if_id);
 174
 175 static struct xfrm_pol_inexact_bin *
 176 xfrm_policy_inexact_lookup_rcu(struct net *net,
 177                                u8 type, u16 family, u8 dir, u32 if_id);
 178 static struct xfrm_policy *
 179 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
 180                         bool excl);
 181 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
 182                                             struct xfrm_policy *policy);
 183
 184 static bool
 185 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
 186                                     struct xfrm_pol_inexact_bin *b,
 187                                     const xfrm_address_t *saddr,
 188                                     const xfrm_address_t *daddr);
 189
 190 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
 191 {
 192         return refcount_inc_not_zero(&policy->refcnt);
 193 }
 194
 195 static inline bool
 196 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 197 {
 198         const struct flowi4 *fl4 = &fl->u.ip4;
 199
 200         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
 201                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
 202                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
 203                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
 204                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
 205                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
 206 }
 207
 208 static inline bool
 209 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 210 {
 211         const struct flowi6 *fl6 = &fl->u.ip6;
 212
 213         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
 214                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
 215                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
 216                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
 217                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
 218                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
 219 }
 220
 221 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
 222                          unsigned short family)
 223 {
 224         switch (family) {
 225         case AF_INET:
 226                 return __xfrm4_selector_match(sel, fl);
 227         case AF_INET6:
 228                 return __xfrm6_selector_match(sel, fl);
 229         }
 230         return false;
 231 }
 232
 233 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 234 {
 235         const struct xfrm_policy_afinfo *afinfo;
 236
 237         if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
 238                 return NULL;
 239         rcu_read_lock();
 240         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 241         if (unlikely(!afinfo))
 242                 rcu_read_unlock();
 243         return afinfo;
 244 }
 245
 246 /* Called with rcu_read_lock(). */
 247 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
 248 {
 249         return rcu_dereference(xfrm_if_cb);
 250 }
 251
 252 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif,
 253                                     const xfrm_address_t *saddr,
 254                                     const xfrm_address_t *daddr,
 255                                     int family, u32 mark)
 256 {
 257         const struct xfrm_policy_afinfo *afinfo;
 258         struct dst_entry *dst;
 259
 260         afinfo = xfrm_policy_get_afinfo(family);
 261         if (unlikely(afinfo == NULL))
 262                 return ERR_PTR(-EAFNOSUPPORT);
 263
 264         dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark);
 265
 266         rcu_read_unlock();
 267
 268         return dst;
 269 }
 270 EXPORT_SYMBOL(__xfrm_dst_lookup);
 271
 272 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
 273                                                 int tos, int oif,
 274                                                 xfrm_address_t *prev_saddr,
 275                                                 xfrm_address_t *prev_daddr,
 276                                                 int family, u32 mark)
 277 {
 278         struct net *net = xs_net(x);
 279         xfrm_address_t *saddr = &x->props.saddr;
 280         xfrm_address_t *daddr = &x->id.daddr;
 281         struct dst_entry *dst;
 282
 283         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 284                 saddr = x->coaddr;
 285                 daddr = prev_daddr;
 286         }
 287         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 288                 saddr = prev_saddr;
 289                 daddr = x->coaddr;
 290         }
 291
 292         dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark);
 293
 294         if (!IS_ERR(dst)) {
 295                 if (prev_saddr != saddr)
 296                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 297                 if (prev_daddr != daddr)
 298                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 299         }
 300
 301         return dst;
 302 }
 303
 304 static inline unsigned long make_jiffies(long secs)
 305 {
 306         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 307                 return MAX_SCHEDULE_TIMEOUT-1;
 308         else
 309                 return secs*HZ;
 310 }
 311
 312 static void xfrm_policy_timer(struct timer_list *t)
 313 {
 314         struct xfrm_policy *xp = from_timer(xp, t, timer);
 315         time64_t now = ktime_get_real_seconds();
 316         time64_t next = TIME64_MAX;
 317         int warn = 0;
 318         int dir;
 319
 320         read_lock(&xp->lock);
 321
 322         if (unlikely(xp->walk.dead))
 323                 goto out;
 324
 325         dir = xfrm_policy_id2dir(xp->index);
 326
 327         if (xp->lft.hard_add_expires_seconds) {
 328                 time64_t tmo = xp->lft.hard_add_expires_seconds +
 329                         xp->curlft.add_time - now;
 330                 if (tmo <= 0)
 331                         goto expired;
 332                 if (tmo < next)
 333                         next = tmo;
 334         }
 335         if (xp->lft.hard_use_expires_seconds) {
 336                 time64_t tmo = xp->lft.hard_use_expires_seconds +
 337                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 338                 if (tmo <= 0)
 339                         goto expired;
 340                 if (tmo < next)
 341                         next = tmo;
 342         }
 343         if (xp->lft.soft_add_expires_seconds) {
 344                 time64_t tmo = xp->lft.soft_add_expires_seconds +
 345                         xp->curlft.add_time - now;
 346                 if (tmo <= 0) {
 347                         warn = 1;
 348                         tmo = XFRM_KM_TIMEOUT;
 349                 }
 350                 if (tmo < next)
 351                         next = tmo;
 352         }
 353         if (xp->lft.soft_use_expires_seconds) {
 354                 time64_t tmo = xp->lft.soft_use_expires_seconds +
 355                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
 356                 if (tmo <= 0) {
 357                         warn = 1;
 358                         tmo = XFRM_KM_TIMEOUT;
 359                 }
 360                 if (tmo < next)
 361                         next = tmo;
 362         }
 363
 364         if (warn)
 365                 km_policy_expired(xp, dir, 0, 0);
 366         if (next != TIME64_MAX &&
 367             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 368                 xfrm_pol_hold(xp);
 369
 370 out:
 371         read_unlock(&xp->lock);
 372         xfrm_pol_put(xp);
 373         return;
 374
 375 expired:
 376         read_unlock(&xp->lock);
 377         if (!xfrm_policy_delete(xp, dir))
 378                 km_policy_expired(xp, dir, 1, 0);
 379         xfrm_pol_put(xp);
 380 }
 381
 382 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 383  * SPD calls.
 384  */
 385
 386 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 387 {
 388         struct xfrm_policy *policy;
 389
 390         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 391
 392         if (policy) {
 393                 write_pnet(&policy->xp_net, net);
 394                 INIT_LIST_HEAD(&policy->walk.all);
 395                 INIT_HLIST_NODE(&policy->bydst_inexact_list);
 396                 INIT_HLIST_NODE(&policy->bydst);
 397                 INIT_HLIST_NODE(&policy->byidx);
 398                 rwlock_init(&policy->lock);
 399                 refcount_set(&policy->refcnt, 1);
 400                 skb_queue_head_init(&policy->polq.hold_queue);
 401                 timer_setup(&policy->timer, xfrm_policy_timer, 0);
 402                 timer_setup(&policy->polq.hold_timer,
 403                             xfrm_policy_queue_process, 0);
 404         }
 405         return policy;
 406 }
 407 EXPORT_SYMBOL(xfrm_policy_alloc);
 408
 409 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
 410 {
 411         struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
 412
 413         security_xfrm_policy_free(policy->security);
 414         kfree(policy);
 415 }
 416
 417 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 418
 419 void xfrm_policy_destroy(struct xfrm_policy *policy)
 420 {
 421         BUG_ON(!policy->walk.dead);
 422
 423         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 424                 BUG();
 425
 426         call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
 427 }
 428 EXPORT_SYMBOL(xfrm_policy_destroy);
 429
 430 /* Rule must be locked. Release descendant resources, announce
 431  * entry dead. The rule must be unlinked from lists to the moment.
 432  */
 433
 434 static void xfrm_policy_kill(struct xfrm_policy *policy)
 435 {
 436         write_lock_bh(&policy->lock);
 437         policy->walk.dead = 1;
 438         write_unlock_bh(&policy->lock);
 439
 440         atomic_inc(&policy->genid);
 441
 442         if (del_timer(&policy->polq.hold_timer))
 443                 xfrm_pol_put(policy);
 444         skb_queue_purge(&policy->polq.hold_queue);
 445
 446         if (del_timer(&policy->timer))
 447                 xfrm_pol_put(policy);
 448
 449         xfrm_pol_put(policy);
 450 }
 451
 452 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 453
 454 static inline unsigned int idx_hash(struct net *net, u32 index)
 455 {
 456         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 457 }
 458
 459 /* calculate policy hash thresholds */
 460 static void __get_hash_thresh(struct net *net,
 461                               unsigned short family, int dir,
 462                               u8 *dbits, u8 *sbits)
 463 {
 464         switch (family) {
 465         case AF_INET:
 466                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 467                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 468                 break;
 469
 470         case AF_INET6:
 471                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 472                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 473                 break;
 474
 475         default:
 476                 *dbits = 0;
 477                 *sbits = 0;
 478         }
 479 }
 480
 481 static struct hlist_head *policy_hash_bysel(struct net *net,
 482                                             const struct xfrm_selector *sel,
 483                                             unsigned short family, int dir)
 484 {
 485         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 486         unsigned int hash;
 487         u8 dbits;
 488         u8 sbits;
 489
 490         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 491         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 492
 493         if (hash == hmask + 1)
 494                 return NULL;
 495
 496         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 497                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 498 }
 499
 500 static struct hlist_head *policy_hash_direct(struct net *net,
 501                                              const xfrm_address_t *daddr,
 502                                              const xfrm_address_t *saddr,
 503                                              unsigned short family, int dir)
 504 {
 505         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 506         unsigned int hash;
 507         u8 dbits;
 508         u8 sbits;
 509
 510         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 511         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 512
 513         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 514                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 515 }
 516
 517 static void xfrm_dst_hash_transfer(struct net *net,
 518                                    struct hlist_head *list,
 519                                    struct hlist_head *ndsttable,
 520                                    unsigned int nhashmask,
 521                                    int dir)
 522 {
 523         struct hlist_node *tmp, *entry0 = NULL;
 524         struct xfrm_policy *pol;
 525         unsigned int h0 = 0;
 526         u8 dbits;
 527         u8 sbits;
 528
 529 redo:
 530         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 531                 unsigned int h;
 532
 533                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 534                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 535                                 pol->family, nhashmask, dbits, sbits);
 536                 if (!entry0) {
 537                         hlist_del_rcu(&pol->bydst);
 538                         hlist_add_head_rcu(&pol->bydst, ndsttable + h);
 539                         h0 = h;
 540                 } else {
 541                         if (h != h0)
 542                                 continue;
 543                         hlist_del_rcu(&pol->bydst);
 544                         hlist_add_behind_rcu(&pol->bydst, entry0);
 545                 }
 546                 entry0 = &pol->bydst;
 547         }
 548         if (!hlist_empty(list)) {
 549                 entry0 = NULL;
 550                 goto redo;
 551         }
 552 }
 553
 554 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 555                                    struct hlist_head *nidxtable,
 556                                    unsigned int nhashmask)
 557 {
 558         struct hlist_node *tmp;
 559         struct xfrm_policy *pol;
 560
 561         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 562                 unsigned int h;
 563
 564                 h = __idx_hash(pol->index, nhashmask);
 565                 hlist_add_head(&pol->byidx, nidxtable+h);
 566         }
 567 }
 568
 569 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 570 {
 571         return ((old_hmask + 1) << 1) - 1;
 572 }
 573
 574 static void xfrm_bydst_resize(struct net *net, int dir)
 575 {
 576         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 577         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 578         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 579         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 580         struct hlist_head *odst;
 581         int i;
 582
 583         if (!ndst)
 584                 return;
 585
 586         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 587         write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
 588
 589         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 590                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 591
 592         for (i = hmask; i >= 0; i--)
 593                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 594
 595         rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
 596         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 597
 598         write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
 599         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 600
 601         synchronize_rcu();
 602
 603         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 604 }
 605
 606 static void xfrm_byidx_resize(struct net *net, int total)
 607 {
 608         unsigned int hmask = net->xfrm.policy_idx_hmask;
 609         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 610         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 611         struct hlist_head *oidx = net->xfrm.policy_byidx;
 612         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 613         int i;
 614
 615         if (!nidx)
 616                 return;
 617
 618         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 619
 620         for (i = hmask; i >= 0; i--)
 621                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 622
 623         net->xfrm.policy_byidx = nidx;
 624         net->xfrm.policy_idx_hmask = nhashmask;
 625
 626         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 627
 628         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 629 }
 630
 631 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 632 {
 633         unsigned int cnt = net->xfrm.policy_count[dir];
 634         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 635
 636         if (total)
 637                 *total += cnt;
 638
 639         if ((hmask + 1) < xfrm_policy_hashmax &&
 640             cnt > hmask)
 641                 return 1;
 642
 643         return 0;
 644 }
 645
 646 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 647 {
 648         unsigned int hmask = net->xfrm.policy_idx_hmask;
 649
 650         if ((hmask + 1) < xfrm_policy_hashmax &&
 651             total > hmask)
 652                 return 1;
 653
 654         return 0;
 655 }
 656
 657 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 658 {
 659         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 660         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 661         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 662         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 663         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 664         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 665         si->spdhcnt = net->xfrm.policy_idx_hmask;
 666         si->spdhmcnt = xfrm_policy_hashmax;
 667 }
 668 EXPORT_SYMBOL(xfrm_spd_getinfo);
 669
 670 static DEFINE_MUTEX(hash_resize_mutex);
 671 static void xfrm_hash_resize(struct work_struct *work)
 672 {
 673         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 674         int dir, total;
 675
 676         mutex_lock(&hash_resize_mutex);
 677
 678         total = 0;
 679         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 680                 if (xfrm_bydst_should_resize(net, dir, &total))
 681                         xfrm_bydst_resize(net, dir);
 682         }
 683         if (xfrm_byidx_should_resize(net, total))
 684                 xfrm_byidx_resize(net, total);
 685
 686         mutex_unlock(&hash_resize_mutex);
 687 }
 688
 689 /* Make sure *pol can be inserted into fastbin.
 690  * Useful to check that later insert requests will be successful
 691  * (provided xfrm_policy_lock is held throughout).
 692  */
 693 static struct xfrm_pol_inexact_bin *
 694 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
 695 {
 696         struct xfrm_pol_inexact_bin *bin, *prev;
 697         struct xfrm_pol_inexact_key k = {
 698                 .family = pol->family,
 699                 .type = pol->type,
 700                 .dir = dir,
 701                 .if_id = pol->if_id,
 702         };
 703         struct net *net = xp_net(pol);
 704
 705         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
 706
 707         write_pnet(&k.net, net);
 708         bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
 709                                      xfrm_pol_inexact_params);
 710         if (bin)
 711                 return bin;
 712
 713         bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
 714         if (!bin)
 715                 return NULL;
 716
 717         bin->k = k;
 718         INIT_HLIST_HEAD(&bin->hhead);
 719         bin->root_d = RB_ROOT;
 720         bin->root_s = RB_ROOT;
 721         seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
 722
 723         prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
 724                                                 &bin->k, &bin->head,
 725                                                 xfrm_pol_inexact_params);
 726         if (!prev) {
 727                 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
 728                 return bin;
 729         }
 730
 731         kfree(bin);
 732
 733         return IS_ERR(prev) ? NULL : prev;
 734 }
 735
 736 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
 737                                                int family, u8 prefixlen)
 738 {
 739         if (xfrm_addr_any(addr, family))
 740                 return true;
 741
 742         if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
 743                 return true;
 744
 745         if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
 746                 return true;
 747
 748         return false;
 749 }
 750
 751 static bool
 752 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
 753 {
 754         const xfrm_address_t *addr;
 755         bool saddr_any, daddr_any;
 756         u8 prefixlen;
 757
 758         addr = &policy->selector.saddr;
 759         prefixlen = policy->selector.prefixlen_s;
 760
 761         saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 762                                                        policy->family,
 763                                                        prefixlen);
 764         addr = &policy->selector.daddr;
 765         prefixlen = policy->selector.prefixlen_d;
 766         daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 767                                                        policy->family,
 768                                                        prefixlen);
 769         return saddr_any && daddr_any;
 770 }
 771
 772 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
 773                                        const xfrm_address_t *addr, u8 prefixlen)
 774 {
 775         node->addr = *addr;
 776         node->prefixlen = prefixlen;
 777 }
 778
 779 static struct xfrm_pol_inexact_node *
 780 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
 781 {
 782         struct xfrm_pol_inexact_node *node;
 783
 784         node = kzalloc(sizeof(*node), GFP_ATOMIC);
 785         if (node)
 786                 xfrm_pol_inexact_node_init(node, addr, prefixlen);
 787
 788         return node;
 789 }
 790
 791 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
 792                                   const xfrm_address_t *b,
 793                                   u8 prefixlen, u16 family)
 794 {
 795         u32 ma, mb, mask;
 796         unsigned int pdw, pbi;
 797         int delta = 0;
 798
 799         switch (family) {
 800         case AF_INET:
 801                 if (prefixlen == 0)
 802                         return 0;
 803                 mask = ~0U << (32 - prefixlen);
 804                 ma = ntohl(a->a4) & mask;
 805                 mb = ntohl(b->a4) & mask;
 806                 if (ma < mb)
 807                         delta = -1;
 808                 else if (ma > mb)
 809                         delta = 1;
 810                 break;
 811         case AF_INET6:
 812                 pdw = prefixlen >> 5;
 813                 pbi = prefixlen & 0x1f;
 814
 815                 if (pdw) {
 816                         delta = memcmp(a->a6, b->a6, pdw << 2);
 817                         if (delta)
 818                                 return delta;
 819                 }
 820                 if (pbi) {
 821                         mask = ~0U << (32 - pbi);
 822                         ma = ntohl(a->a6[pdw]) & mask;
 823                         mb = ntohl(b->a6[pdw]) & mask;
 824                         if (ma < mb)
 825                                 delta = -1;
 826                         else if (ma > mb)
 827                                 delta = 1;
 828                 }
 829                 break;
 830         default:
 831                 break;
 832         }
 833
 834         return delta;
 835 }
 836
 837 static void xfrm_policy_inexact_list_reinsert(struct net *net,
 838                                               struct xfrm_pol_inexact_node *n,
 839                                               u16 family)
 840 {
 841         unsigned int matched_s, matched_d;
 842         struct xfrm_policy *policy, *p;
 843
 844         matched_s = 0;
 845         matched_d = 0;
 846
 847         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 848                 struct hlist_node *newpos = NULL;
 849                 bool matches_s, matches_d;
 850
 851                 if (!policy->bydst_reinsert)
 852                         continue;
 853
 854                 WARN_ON_ONCE(policy->family != family);
 855
 856                 policy->bydst_reinsert = false;
 857                 hlist_for_each_entry(p, &n->hhead, bydst) {
 858                         if (policy->priority > p->priority)
 859                                 newpos = &p->bydst;
 860                         else if (policy->priority == p->priority &&
 861                                  policy->pos > p->pos)
 862                                 newpos = &p->bydst;
 863                         else
 864                                 break;
 865                 }
 866
 867                 if (newpos)
 868                         hlist_add_behind_rcu(&policy->bydst, newpos);
 869                 else
 870                         hlist_add_head_rcu(&policy->bydst, &n->hhead);
 871
 872                 /* paranoia checks follow.
 873                  * Check that the reinserted policy matches at least
 874                  * saddr or daddr for current node prefix.
 875                  *
 876                  * Matching both is fine, matching saddr in one policy
 877                  * (but not daddr) and then matching only daddr in another
 878                  * is a bug.
 879                  */
 880                 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
 881                                                    &n->addr,
 882                                                    n->prefixlen,
 883                                                    family) == 0;
 884                 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
 885                                                    &n->addr,
 886                                                    n->prefixlen,
 887                                                    family) == 0;
 888                 if (matches_s && matches_d)
 889                         continue;
 890
 891                 WARN_ON_ONCE(!matches_s && !matches_d);
 892                 if (matches_s)
 893                         matched_s++;
 894                 if (matches_d)
 895                         matched_d++;
 896                 WARN_ON_ONCE(matched_s && matched_d);
 897         }
 898 }
 899
 900 static void xfrm_policy_inexact_node_reinsert(struct net *net,
 901                                               struct xfrm_pol_inexact_node *n,
 902                                               struct rb_root *new,
 903                                               u16 family)
 904 {
 905         struct xfrm_pol_inexact_node *node;
 906         struct rb_node **p, *parent;
 907
 908         /* we should not have another subtree here */
 909         WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
 910 restart:
 911         parent = NULL;
 912         p = &new->rb_node;
 913         while (*p) {
 914                 u8 prefixlen;
 915                 int delta;
 916
 917                 parent = *p;
 918                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
 919
 920                 prefixlen = min(node->prefixlen, n->prefixlen);
 921
 922                 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
 923                                                prefixlen, family);
 924                 if (delta < 0) {
 925                         p = &parent->rb_left;
 926                 } else if (delta > 0) {
 927                         p = &parent->rb_right;
 928                 } else {
 929                         bool same_prefixlen = node->prefixlen == n->prefixlen;
 930                         struct xfrm_policy *tmp;
 931
 932                         hlist_for_each_entry(tmp, &n->hhead, bydst) {
 933                                 tmp->bydst_reinsert = true;
 934                                 hlist_del_rcu(&tmp->bydst);
 935                         }
 936
 937                         node->prefixlen = prefixlen;
 938
 939                         xfrm_policy_inexact_list_reinsert(net, node, family);
 940
 941                         if (same_prefixlen) {
 942                                 kfree_rcu(n, rcu);
 943                                 return;
 944                         }
 945
 946                         rb_erase(*p, new);
 947                         kfree_rcu(n, rcu);
 948                         n = node;
 949                         goto restart;
 950                 }
 951         }
 952
 953         rb_link_node_rcu(&n->node, parent, p);
 954         rb_insert_color(&n->node, new);
 955 }
 956
 957 /* merge nodes v and n */
 958 static void xfrm_policy_inexact_node_merge(struct net *net,
 959                                            struct xfrm_pol_inexact_node *v,
 960                                            struct xfrm_pol_inexact_node *n,
 961                                            u16 family)
 962 {
 963         struct xfrm_pol_inexact_node *node;
 964         struct xfrm_policy *tmp;
 965         struct rb_node *rnode;
 966
 967         /* To-be-merged node v has a subtree.
 968          *
 969          * Dismantle it and insert its nodes to n->root.
 970          */
 971         while ((rnode = rb_first(&v->root)) != NULL) {
 972                 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
 973                 rb_erase(&node->node, &v->root);
 974                 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
 975                                                   family);
 976         }
 977
 978         hlist_for_each_entry(tmp, &v->hhead, bydst) {
 979                 tmp->bydst_reinsert = true;
 980                 hlist_del_rcu(&tmp->bydst);
 981         }
 982
 983         xfrm_policy_inexact_list_reinsert(net, n, family);
 984 }
 985
 986 static struct xfrm_pol_inexact_node *
 987 xfrm_policy_inexact_insert_node(struct net *net,
 988                                 struct rb_root *root,
 989                                 xfrm_address_t *addr,
 990                                 u16 family, u8 prefixlen, u8 dir)
 991 {
 992         struct xfrm_pol_inexact_node *cached = NULL;
 993         struct rb_node **p, *parent = NULL;
 994         struct xfrm_pol_inexact_node *node;
 995
 996         p = &root->rb_node;
 997         while (*p) {
 998                 int delta;
 999
1000                 parent = *p;
1001                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1002
1003                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1004                                                node->prefixlen,
1005                                                family);
1006                 if (delta == 0 && prefixlen >= node->prefixlen) {
1007                         WARN_ON_ONCE(cached); /* ipsec policies got lost */
1008                         return node;
1009                 }
1010
1011                 if (delta < 0)
1012                         p = &parent->rb_left;
1013                 else
1014                         p = &parent->rb_right;
1015
1016                 if (prefixlen < node->prefixlen) {
1017                         delta = xfrm_policy_addr_delta(addr, &node->addr,
1018                                                        prefixlen,
1019                                                        family);
1020                         if (delta)
1021                                 continue;
1022
1023                         /* This node is a subnet of the new prefix. It needs
1024                          * to be removed and re-inserted with the smaller
1025                          * prefix and all nodes that are now also covered
1026                          * by the reduced prefixlen.
1027                          */
1028                         rb_erase(&node->node, root);
1029
1030                         if (!cached) {
1031                                 xfrm_pol_inexact_node_init(node, addr,
1032                                                            prefixlen);
1033                                 cached = node;
1034                         } else {
1035                                 /* This node also falls within the new
1036                                  * prefixlen. Merge the to-be-reinserted
1037                                  * node and this one.
1038                                  */
1039                                 xfrm_policy_inexact_node_merge(net, node,
1040                                                                cached, family);
1041                                 kfree_rcu(node, rcu);
1042                         }
1043
1044                         /* restart */
1045                         p = &root->rb_node;
1046                         parent = NULL;
1047                 }
1048         }
1049
1050         node = cached;
1051         if (!node) {
1052                 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1053                 if (!node)
1054                         return NULL;
1055         }
1056
1057         rb_link_node_rcu(&node->node, parent, p);
1058         rb_insert_color(&node->node, root);
1059
1060         return node;
1061 }
1062
1063 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1064 {
1065         struct xfrm_pol_inexact_node *node;
1066         struct rb_node *rn = rb_first(r);
1067
1068         while (rn) {
1069                 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1070
1071                 xfrm_policy_inexact_gc_tree(&node->root, rm);
1072                 rn = rb_next(rn);
1073
1074                 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1075                         WARN_ON_ONCE(rm);
1076                         continue;
1077                 }
1078
1079                 rb_erase(&node->node, r);
1080                 kfree_rcu(node, rcu);
1081         }
1082 }
1083
1084 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1085 {
1086         write_seqcount_begin(&b->count);
1087         xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1088         xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1089         write_seqcount_end(&b->count);
1090
1091         if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1092             !hlist_empty(&b->hhead)) {
1093                 WARN_ON_ONCE(net_exit);
1094                 return;
1095         }
1096
1097         if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1098                                    xfrm_pol_inexact_params) == 0) {
1099                 list_del(&b->inexact_bins);
1100                 kfree_rcu(b, rcu);
1101         }
1102 }
1103
1104 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1105 {
1106         struct net *net = read_pnet(&b->k.net);
1107
1108         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1109         __xfrm_policy_inexact_prune_bin(b, false);
1110         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1111 }
1112
1113 static void __xfrm_policy_inexact_flush(struct net *net)
1114 {
1115         struct xfrm_pol_inexact_bin *bin, *t;
1116
1117         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1118
1119         list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1120                 __xfrm_policy_inexact_prune_bin(bin, false);
1121 }
1122
1123 static struct hlist_head *
1124 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1125                                 struct xfrm_policy *policy, u8 dir)
1126 {
1127         struct xfrm_pol_inexact_node *n;
1128         struct net *net;
1129
1130         net = xp_net(policy);
1131         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1132
1133         if (xfrm_policy_inexact_insert_use_any_list(policy))
1134                 return &bin->hhead;
1135
1136         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1137                                                policy->family,
1138                                                policy->selector.prefixlen_d)) {
1139                 write_seqcount_begin(&bin->count);
1140                 n = xfrm_policy_inexact_insert_node(net,
1141                                                     &bin->root_s,
1142                                                     &policy->selector.saddr,
1143                                                     policy->family,
1144                                                     policy->selector.prefixlen_s,
1145                                                     dir);
1146                 write_seqcount_end(&bin->count);
1147                 if (!n)
1148                         return NULL;
1149
1150                 return &n->hhead;
1151         }
1152
1153         /* daddr is fixed */
1154         write_seqcount_begin(&bin->count);
1155         n = xfrm_policy_inexact_insert_node(net,
1156                                             &bin->root_d,
1157                                             &policy->selector.daddr,
1158                                             policy->family,
1159                                             policy->selector.prefixlen_d, dir);
1160         write_seqcount_end(&bin->count);
1161         if (!n)
1162                 return NULL;
1163
1164         /* saddr is wildcard */
1165         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1166                                                policy->family,
1167                                                policy->selector.prefixlen_s))
1168                 return &n->hhead;
1169
1170         write_seqcount_begin(&bin->count);
1171         n = xfrm_policy_inexact_insert_node(net,
1172                                             &n->root,
1173                                             &policy->selector.saddr,
1174                                             policy->family,
1175                                             policy->selector.prefixlen_s, dir);
1176         write_seqcount_end(&bin->count);
1177         if (!n)
1178                 return NULL;
1179
1180         return &n->hhead;
1181 }
1182
1183 static struct xfrm_policy *
1184 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1185 {
1186         struct xfrm_pol_inexact_bin *bin;
1187         struct xfrm_policy *delpol;
1188         struct hlist_head *chain;
1189         struct net *net;
1190
1191         bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1192         if (!bin)
1193                 return ERR_PTR(-ENOMEM);
1194
1195         net = xp_net(policy);
1196         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1197
1198         chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1199         if (!chain) {
1200                 __xfrm_policy_inexact_prune_bin(bin, false);
1201                 return ERR_PTR(-ENOMEM);
1202         }
1203
1204         delpol = xfrm_policy_insert_list(chain, policy, excl);
1205         if (delpol && excl) {
1206                 __xfrm_policy_inexact_prune_bin(bin, false);
1207                 return ERR_PTR(-EEXIST);
1208         }
1209
1210         chain = &net->xfrm.policy_inexact[dir];
1211         xfrm_policy_insert_inexact_list(chain, policy);
1212
1213         if (delpol)
1214                 __xfrm_policy_inexact_prune_bin(bin, false);
1215
1216         return delpol;
1217 }
1218
1219 static void xfrm_hash_rebuild(struct work_struct *work)
1220 {
1221         struct net *net = container_of(work, struct net,
1222                                        xfrm.policy_hthresh.work);
1223         unsigned int hmask;
1224         struct xfrm_policy *pol;
1225         struct xfrm_policy *policy;
1226         struct hlist_head *chain;
1227         struct hlist_head *odst;
1228         struct hlist_node *newpos;
1229         int i;
1230         int dir;
1231         unsigned seq;
1232         u8 lbits4, rbits4, lbits6, rbits6;
1233
1234         mutex_lock(&hash_resize_mutex);
1235
1236         /* read selector prefixlen thresholds */
1237         do {
1238                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1239
1240                 lbits4 = net->xfrm.policy_hthresh.lbits4;
1241                 rbits4 = net->xfrm.policy_hthresh.rbits4;
1242                 lbits6 = net->xfrm.policy_hthresh.lbits6;
1243                 rbits6 = net->xfrm.policy_hthresh.rbits6;
1244         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1245
1246         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1247         write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
1248
1249         /* make sure that we can insert the indirect policies again before
1250          * we start with destructive action.
1251          */
1252         list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1253                 struct xfrm_pol_inexact_bin *bin;
1254                 u8 dbits, sbits;
1255
1256                 dir = xfrm_policy_id2dir(policy->index);
1257                 if (policy->walk.dead || dir >= XFRM_POLICY_MAX)
1258                         continue;
1259
1260                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1261                         if (policy->family == AF_INET) {
1262                                 dbits = rbits4;
1263                                 sbits = lbits4;
1264                         } else {
1265                                 dbits = rbits6;
1266                                 sbits = lbits6;
1267                         }
1268                 } else {
1269                         if (policy->family == AF_INET) {
1270                                 dbits = lbits4;
1271                                 sbits = rbits4;
1272                         } else {
1273                                 dbits = lbits6;
1274                                 sbits = rbits6;
1275                         }
1276                 }
1277
1278                 if (policy->selector.prefixlen_d < dbits ||
1279                     policy->selector.prefixlen_s < sbits)
1280                         continue;
1281
1282                 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1283                 if (!bin)
1284                         goto out_unlock;
1285
1286                 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1287                         goto out_unlock;
1288         }
1289
1290         /* reset the bydst and inexact table in all directions */
1291         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1292                 struct hlist_node *n;
1293
1294                 hlist_for_each_entry_safe(policy, n,
1295                                           &net->xfrm.policy_inexact[dir],
1296                                           bydst_inexact_list) {
1297                         hlist_del_rcu(&policy->bydst);
1298                         hlist_del_init(&policy->bydst_inexact_list);
1299                 }
1300
1301                 hmask = net->xfrm.policy_bydst[dir].hmask;
1302                 odst = net->xfrm.policy_bydst[dir].table;
1303                 for (i = hmask; i >= 0; i--) {
1304                         hlist_for_each_entry_safe(policy, n, odst + i, bydst)
1305                                 hlist_del_rcu(&policy->bydst);
1306                 }
1307                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1308                         /* dir out => dst = remote, src = local */
1309                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1310                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1311                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1312                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1313                 } else {
1314                         /* dir in/fwd => dst = local, src = remote */
1315                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1316                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1317                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1318                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1319                 }
1320         }
1321
1322         /* re-insert all policies by order of creation */
1323         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1324                 if (policy->walk.dead)
1325                         continue;
1326                 dir = xfrm_policy_id2dir(policy->index);
1327                 if (dir >= XFRM_POLICY_MAX) {
1328                         /* skip socket policies */
1329                         continue;
1330                 }
1331                 newpos = NULL;
1332                 chain = policy_hash_bysel(net, &policy->selector,
1333                                           policy->family, dir);
1334
1335                 if (!chain) {
1336                         void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1337
1338                         WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1339                         continue;
1340                 }
1341
1342                 hlist_for_each_entry(pol, chain, bydst) {
1343                         if (policy->priority >= pol->priority)
1344                                 newpos = &pol->bydst;
1345                         else
1346                                 break;
1347                 }
1348                 if (newpos)
1349                         hlist_add_behind_rcu(&policy->bydst, newpos);
1350                 else
1351                         hlist_add_head_rcu(&policy->bydst, chain);
1352         }
1353
1354 out_unlock:
1355         __xfrm_policy_inexact_flush(net);
1356         write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1357         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1358
1359         mutex_unlock(&hash_resize_mutex);
1360 }
1361
1362 void xfrm_policy_hash_rebuild(struct net *net)
1363 {
1364         schedule_work(&net->xfrm.policy_hthresh.work);
1365 }
1366 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1367
1368 /* Generate new index... KAME seems to generate them ordered by cost
1369  * of an absolute inpredictability of ordering of rules. This will not pass. */
1370 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1371 {
1372         static u32 idx_generator;
1373
1374         for (;;) {
1375                 struct hlist_head *list;
1376                 struct xfrm_policy *p;
1377                 u32 idx;
1378                 int found;
1379
1380                 if (!index) {
1381                         idx = (idx_generator | dir);
1382                         idx_generator += 8;
1383                 } else {
1384                         idx = index;
1385                         index = 0;
1386                 }
1387
1388                 if (idx == 0)
1389                         idx = 8;
1390                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1391                 found = 0;
1392                 hlist_for_each_entry(p, list, byidx) {
1393                         if (p->index == idx) {
1394                                 found = 1;
1395                                 break;
1396                         }
1397                 }
1398                 if (!found)
1399                         return idx;
1400         }
1401 }
1402
1403 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1404 {
1405         u32 *p1 = (u32 *) s1;
1406         u32 *p2 = (u32 *) s2;
1407         int len = sizeof(struct xfrm_selector) / sizeof(u32);
1408         int i;
1409
1410         for (i = 0; i < len; i++) {
1411                 if (p1[i] != p2[i])
1412                         return 1;
1413         }
1414
1415         return 0;
1416 }
1417
1418 static void xfrm_policy_requeue(struct xfrm_policy *old,
1419                                 struct xfrm_policy *new)
1420 {
1421         struct xfrm_policy_queue *pq = &old->polq;
1422         struct sk_buff_head list;
1423
1424         if (skb_queue_empty(&pq->hold_queue))
1425                 return;
1426
1427         __skb_queue_head_init(&list);
1428
1429         spin_lock_bh(&pq->hold_queue.lock);
1430         skb_queue_splice_init(&pq->hold_queue, &list);
1431         if (del_timer(&pq->hold_timer))
1432                 xfrm_pol_put(old);
1433         spin_unlock_bh(&pq->hold_queue.lock);
1434
1435         pq = &new->polq;
1436
1437         spin_lock_bh(&pq->hold_queue.lock);
1438         skb_queue_splice(&list, &pq->hold_queue);
1439         pq->timeout = XFRM_QUEUE_TMO_MIN;
1440         if (!mod_timer(&pq->hold_timer, jiffies))
1441                 xfrm_pol_hold(new);
1442         spin_unlock_bh(&pq->hold_queue.lock);
1443 }
1444
1445 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1446                                           struct xfrm_policy *pol)
1447 {
1448         return mark->v == pol->mark.v && mark->m == pol->mark.m;
1449 }
1450
1451 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1452 {
1453         const struct xfrm_pol_inexact_key *k = data;
1454         u32 a = k->type << 24 | k->dir << 16 | k->family;
1455
1456         return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1457                             seed);
1458 }
1459
1460 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1461 {
1462         const struct xfrm_pol_inexact_bin *b = data;
1463
1464         return xfrm_pol_bin_key(&b->k, 0, seed);
1465 }
1466
1467 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1468                             const void *ptr)
1469 {
1470         const struct xfrm_pol_inexact_key *key = arg->key;
1471         const struct xfrm_pol_inexact_bin *b = ptr;
1472         int ret;
1473
1474         if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1475                 return -1;
1476
1477         ret = b->k.dir ^ key->dir;
1478         if (ret)
1479                 return ret;
1480
1481         ret = b->k.type ^ key->type;
1482         if (ret)
1483                 return ret;
1484
1485         ret = b->k.family ^ key->family;
1486         if (ret)
1487                 return ret;
1488
1489         return b->k.if_id ^ key->if_id;
1490 }
1491
1492 static const struct rhashtable_params xfrm_pol_inexact_params = {
1493         .head_offset            = offsetof(struct xfrm_pol_inexact_bin, head),
1494         .hashfn                 = xfrm_pol_bin_key,
1495         .obj_hashfn             = xfrm_pol_bin_obj,
1496         .obj_cmpfn              = xfrm_pol_bin_cmp,
1497         .automatic_shrinking    = true,
1498 };
1499
1500 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain,
1501                                             struct xfrm_policy *policy)
1502 {
1503         struct xfrm_policy *pol, *delpol = NULL;
1504         struct hlist_node *newpos = NULL;
1505         int i = 0;
1506
1507         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1508                 if (pol->type == policy->type &&
1509                     pol->if_id == policy->if_id &&
1510                     !selector_cmp(&pol->selector, &policy->selector) &&
1511                     xfrm_policy_mark_match(&policy->mark, pol) &&
1512                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1513                     !WARN_ON(delpol)) {
1514                         delpol = pol;
1515                         if (policy->priority > pol->priority)
1516                                 continue;
1517                 } else if (policy->priority >= pol->priority) {
1518                         newpos = &pol->bydst_inexact_list;
1519                         continue;
1520                 }
1521                 if (delpol)
1522                         break;
1523         }
1524
1525         if (newpos)
1526                 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos);
1527         else
1528                 hlist_add_head_rcu(&policy->bydst_inexact_list, chain);
1529
1530         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
1531                 pol->pos = i;
1532                 i++;
1533         }
1534 }
1535
1536 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1537                                                    struct xfrm_policy *policy,
1538                                                    bool excl)
1539 {
1540         struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1541
1542         hlist_for_each_entry(pol, chain, bydst) {
1543                 if (pol->type == policy->type &&
1544                     pol->if_id == policy->if_id &&
1545                     !selector_cmp(&pol->selector, &policy->selector) &&
1546                     xfrm_policy_mark_match(&policy->mark, pol) &&
1547                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1548                     !WARN_ON(delpol)) {
1549                         if (excl)
1550                                 return ERR_PTR(-EEXIST);
1551                         delpol = pol;
1552                         if (policy->priority > pol->priority)
1553                                 continue;
1554                 } else if (policy->priority >= pol->priority) {
1555                         newpos = pol;
1556                         continue;
1557                 }
1558                 if (delpol)
1559                         break;
1560         }
1561
1562         if (newpos)
1563                 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1564         else
1565                 hlist_add_head_rcu(&policy->bydst, chain);
1566
1567         return delpol;
1568 }
1569
1570 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1571 {
1572         struct net *net = xp_net(policy);
1573         struct xfrm_policy *delpol;
1574         struct hlist_head *chain;
1575
1576         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1577         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1578         if (chain)
1579                 delpol = xfrm_policy_insert_list(chain, policy, excl);
1580         else
1581                 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1582
1583         if (IS_ERR(delpol)) {
1584                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1585                 return PTR_ERR(delpol);
1586         }
1587
1588         __xfrm_policy_link(policy, dir);
1589
1590         /* After previous checking, family can either be AF_INET or AF_INET6 */
1591         if (policy->family == AF_INET)
1592                 rt_genid_bump_ipv4(net);
1593         else
1594                 rt_genid_bump_ipv6(net);
1595
1596         if (delpol) {
1597                 xfrm_policy_requeue(delpol, policy);
1598                 __xfrm_policy_unlink(delpol, dir);
1599         }
1600         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1601         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1602         policy->curlft.add_time = ktime_get_real_seconds();
1603         policy->curlft.use_time = 0;
1604         if (!mod_timer(&policy->timer, jiffies + HZ))
1605                 xfrm_pol_hold(policy);
1606         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1607
1608         if (delpol)
1609                 xfrm_policy_kill(delpol);
1610         else if (xfrm_bydst_should_resize(net, dir, NULL))
1611                 schedule_work(&net->xfrm.policy_hash_work);
1612
1613         return 0;
1614 }
1615 EXPORT_SYMBOL(xfrm_policy_insert);
1616
1617 static struct xfrm_policy *
1618 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1619                         u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1620                         struct xfrm_sec_ctx *ctx)
1621 {
1622         struct xfrm_policy *pol;
1623
1624         if (!chain)
1625                 return NULL;
1626
1627         hlist_for_each_entry(pol, chain, bydst) {
1628                 if (pol->type == type &&
1629                     pol->if_id == if_id &&
1630                     xfrm_policy_mark_match(mark, pol) &&
1631                     !selector_cmp(sel, &pol->selector) &&
1632                     xfrm_sec_ctx_match(ctx, pol->security))
1633                         return pol;
1634         }
1635
1636         return NULL;
1637 }
1638
1639 struct xfrm_policy *
1640 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1641                       u8 type, int dir, struct xfrm_selector *sel,
1642                       struct xfrm_sec_ctx *ctx, int delete, int *err)
1643 {
1644         struct xfrm_pol_inexact_bin *bin = NULL;
1645         struct xfrm_policy *pol, *ret = NULL;
1646         struct hlist_head *chain;
1647
1648         *err = 0;
1649         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1650         chain = policy_hash_bysel(net, sel, sel->family, dir);
1651         if (!chain) {
1652                 struct xfrm_pol_inexact_candidates cand;
1653                 int i;
1654
1655                 bin = xfrm_policy_inexact_lookup(net, type,
1656                                                  sel->family, dir, if_id);
1657                 if (!bin) {
1658                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1659                         return NULL;
1660                 }
1661
1662                 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1663                                                          &sel->saddr,
1664                                                          &sel->daddr)) {
1665                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1666                         return NULL;
1667                 }
1668
1669                 pol = NULL;
1670                 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1671                         struct xfrm_policy *tmp;
1672
1673                         tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1674                                                       if_id, type, dir,
1675                                                       sel, ctx);
1676                         if (!tmp)
1677                                 continue;
1678
1679                         if (!pol || tmp->pos < pol->pos)
1680                                 pol = tmp;
1681                 }
1682         } else {
1683                 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1684                                               sel, ctx);
1685         }
1686
1687         if (pol) {
1688                 xfrm_pol_hold(pol);
1689                 if (delete) {
1690                         *err = security_xfrm_policy_delete(pol->security);
1691                         if (*err) {
1692                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1693                                 return pol;
1694                         }
1695                         __xfrm_policy_unlink(pol, dir);
1696                 }
1697                 ret = pol;
1698         }
1699         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1700
1701         if (ret && delete)
1702                 xfrm_policy_kill(ret);
1703         if (bin && delete)
1704                 xfrm_policy_inexact_prune_bin(bin);
1705         return ret;
1706 }
1707 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1708
1709 struct xfrm_policy *
1710 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1711                  u8 type, int dir, u32 id, int delete, int *err)
1712 {
1713         struct xfrm_policy *pol, *ret;
1714         struct hlist_head *chain;
1715
1716         *err = -ENOENT;
1717         if (xfrm_policy_id2dir(id) != dir)
1718                 return NULL;
1719
1720         *err = 0;
1721         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1722         chain = net->xfrm.policy_byidx + idx_hash(net, id);
1723         ret = NULL;
1724         hlist_for_each_entry(pol, chain, byidx) {
1725                 if (pol->type == type && pol->index == id &&
1726                     pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1727                         xfrm_pol_hold(pol);
1728                         if (delete) {
1729                                 *err = security_xfrm_policy_delete(
1730                                                                 pol->security);
1731                                 if (*err) {
1732                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1733                                         return pol;
1734                                 }
1735                                 __xfrm_policy_unlink(pol, dir);
1736                         }
1737                         ret = pol;
1738                         break;
1739                 }
1740         }
1741         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1742
1743         if (ret && delete)
1744                 xfrm_policy_kill(ret);
1745         return ret;
1746 }
1747 EXPORT_SYMBOL(xfrm_policy_byid);
1748
1749 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1750 static inline int
1751 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1752 {
1753         struct xfrm_policy *pol;
1754         int err = 0;
1755
1756         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1757                 if (pol->walk.dead ||
1758                     xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1759                     pol->type != type)
1760                         continue;
1761
1762                 err = security_xfrm_policy_delete(pol->security);
1763                 if (err) {
1764                         xfrm_audit_policy_delete(pol, 0, task_valid);
1765                         return err;
1766                 }
1767         }
1768         return err;
1769 }
1770 #else
1771 static inline int
1772 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1773 {
1774         return 0;
1775 }
1776 #endif
1777
1778 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1779 {
1780         int dir, err = 0, cnt = 0;
1781         struct xfrm_policy *pol;
1782
1783         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1784
1785         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1786         if (err)
1787                 goto out;
1788
1789 again:
1790         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1791                 dir = xfrm_policy_id2dir(pol->index);
1792                 if (pol->walk.dead ||
1793                     dir >= XFRM_POLICY_MAX ||
1794                     pol->type != type)
1795                         continue;
1796
1797                 __xfrm_policy_unlink(pol, dir);
1798                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1799                 cnt++;
1800                 xfrm_audit_policy_delete(pol, 1, task_valid);
1801                 xfrm_policy_kill(pol);
1802                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1803                 goto again;
1804         }
1805         if (cnt)
1806                 __xfrm_policy_inexact_flush(net);
1807         else
1808                 err = -ESRCH;
1809 out:
1810         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1811         return err;
1812 }
1813 EXPORT_SYMBOL(xfrm_policy_flush);
1814
1815 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1816                      int (*func)(struct xfrm_policy *, int, int, void*),
1817                      void *data)
1818 {
1819         struct xfrm_policy *pol;
1820         struct xfrm_policy_walk_entry *x;
1821         int error = 0;
1822
1823         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1824             walk->type != XFRM_POLICY_TYPE_ANY)
1825                 return -EINVAL;
1826
1827         if (list_empty(&walk->walk.all) && walk->seq != 0)
1828                 return 0;
1829
1830         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1831         if (list_empty(&walk->walk.all))
1832                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1833         else
1834                 x = list_first_entry(&walk->walk.all,
1835                                      struct xfrm_policy_walk_entry, all);
1836
1837         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1838                 if (x->dead)
1839                         continue;
1840                 pol = container_of(x, struct xfrm_policy, walk);
1841                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1842                     walk->type != pol->type)
1843                         continue;
1844                 error = func(pol, xfrm_policy_id2dir(pol->index),
1845                              walk->seq, data);
1846                 if (error) {
1847                         list_move_tail(&walk->walk.all, &x->all);
1848                         goto out;
1849                 }
1850                 walk->seq++;
1851         }
1852         if (walk->seq == 0) {
1853                 error = -ENOENT;
1854                 goto out;
1855         }
1856         list_del_init(&walk->walk.all);
1857 out:
1858         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1859         return error;
1860 }
1861 EXPORT_SYMBOL(xfrm_policy_walk);
1862
1863 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1864 {
1865         INIT_LIST_HEAD(&walk->walk.all);
1866         walk->walk.dead = 1;
1867         walk->type = type;
1868         walk->seq = 0;
1869 }
1870 EXPORT_SYMBOL(xfrm_policy_walk_init);
1871
1872 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1873 {
1874         if (list_empty(&walk->walk.all))
1875                 return;
1876
1877         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1878         list_del(&walk->walk.all);
1879         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1880 }
1881 EXPORT_SYMBOL(xfrm_policy_walk_done);
1882
1883 /*
1884  * Find policy to apply to this flow.
1885  *
1886  * Returns 0 if policy found, else an -errno.
1887  */
1888 static int xfrm_policy_match(const struct xfrm_policy *pol,
1889                              const struct flowi *fl,
1890                              u8 type, u16 family, int dir, u32 if_id)
1891 {
1892         const struct xfrm_selector *sel = &pol->selector;
1893         int ret = -ESRCH;
1894         bool match;
1895
1896         if (pol->family != family ||
1897             pol->if_id != if_id ||
1898             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1899             pol->type != type)
1900                 return ret;
1901
1902         match = xfrm_selector_match(sel, fl, family);
1903         if (match)
1904                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
1905         return ret;
1906 }
1907
1908 static struct xfrm_pol_inexact_node *
1909 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1910                                 seqcount_spinlock_t *count,
1911                                 const xfrm_address_t *addr, u16 family)
1912 {
1913         const struct rb_node *parent;
1914         int seq;
1915
1916 again:
1917         seq = read_seqcount_begin(count);
1918
1919         parent = rcu_dereference_raw(r->rb_node);
1920         while (parent) {
1921                 struct xfrm_pol_inexact_node *node;
1922                 int delta;
1923
1924                 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
1925
1926                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1927                                                node->prefixlen, family);
1928                 if (delta < 0) {
1929                         parent = rcu_dereference_raw(parent->rb_left);
1930                         continue;
1931                 } else if (delta > 0) {
1932                         parent = rcu_dereference_raw(parent->rb_right);
1933                         continue;
1934                 }
1935
1936                 return node;
1937         }
1938
1939         if (read_seqcount_retry(count, seq))
1940                 goto again;
1941
1942         return NULL;
1943 }
1944
1945 static bool
1946 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
1947                                     struct xfrm_pol_inexact_bin *b,
1948                                     const xfrm_address_t *saddr,
1949                                     const xfrm_address_t *daddr)
1950 {
1951         struct xfrm_pol_inexact_node *n;
1952         u16 family;
1953
1954         if (!b)
1955                 return false;
1956
1957         family = b->k.family;
1958         memset(cand, 0, sizeof(*cand));
1959         cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
1960
1961         n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
1962                                             family);
1963         if (n) {
1964                 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
1965                 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
1966                                                     family);
1967                 if (n)
1968                         cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
1969         }
1970
1971         n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
1972                                             family);
1973         if (n)
1974                 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
1975
1976         return true;
1977 }
1978
1979 static struct xfrm_pol_inexact_bin *
1980 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
1981                                u8 dir, u32 if_id)
1982 {
1983         struct xfrm_pol_inexact_key k = {
1984                 .family = family,
1985                 .type = type,
1986                 .dir = dir,
1987                 .if_id = if_id,
1988         };
1989
1990         write_pnet(&k.net, net);
1991
1992         return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
1993                                  xfrm_pol_inexact_params);
1994 }
1995
1996 static struct xfrm_pol_inexact_bin *
1997 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
1998                            u8 dir, u32 if_id)
1999 {
2000         struct xfrm_pol_inexact_bin *bin;
2001
2002         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2003
2004         rcu_read_lock();
2005         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2006         rcu_read_unlock();
2007
2008         return bin;
2009 }
2010
2011 static struct xfrm_policy *
2012 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2013                               struct xfrm_policy *prefer,
2014                               const struct flowi *fl,
2015                               u8 type, u16 family, int dir, u32 if_id)
2016 {
2017         u32 priority = prefer ? prefer->priority : ~0u;
2018         struct xfrm_policy *pol;
2019
2020         if (!chain)
2021                 return NULL;
2022
2023         hlist_for_each_entry_rcu(pol, chain, bydst) {
2024                 int err;
2025
2026                 if (pol->priority > priority)
2027                         break;
2028
2029                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2030                 if (err) {
2031                         if (err != -ESRCH)
2032                                 return ERR_PTR(err);
2033
2034                         continue;
2035                 }
2036
2037                 if (prefer) {
2038                         /* matches.  Is it older than *prefer? */
2039                         if (pol->priority == priority &&
2040                             prefer->pos < pol->pos)
2041                                 return prefer;
2042                 }
2043
2044                 return pol;
2045         }
2046
2047         return NULL;
2048 }
2049
2050 static struct xfrm_policy *
2051 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2052                             struct xfrm_policy *prefer,
2053                             const struct flowi *fl,
2054                             u8 type, u16 family, int dir, u32 if_id)
2055 {
2056         struct xfrm_policy *tmp;
2057         int i;
2058
2059         for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2060                 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2061                                                     prefer,
2062                                                     fl, type, family, dir,
2063                                                     if_id);
2064                 if (!tmp)
2065                         continue;
2066
2067                 if (IS_ERR(tmp))
2068                         return tmp;
2069                 prefer = tmp;
2070         }
2071
2072         return prefer;
2073 }
2074
2075 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2076                                                      const struct flowi *fl,
2077                                                      u16 family, u8 dir,
2078                                                      u32 if_id)
2079 {
2080         struct xfrm_pol_inexact_candidates cand;
2081         const xfrm_address_t *daddr, *saddr;
2082         struct xfrm_pol_inexact_bin *bin;
2083         struct xfrm_policy *pol, *ret;
2084         struct hlist_head *chain;
2085         unsigned int sequence;
2086         int err;
2087
2088         daddr = xfrm_flowi_daddr(fl, family);
2089         saddr = xfrm_flowi_saddr(fl, family);
2090         if (unlikely(!daddr || !saddr))
2091                 return NULL;
2092
2093         rcu_read_lock();
2094  retry:
2095         do {
2096                 sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2097                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2098         } while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2099
2100         ret = NULL;
2101         hlist_for_each_entry_rcu(pol, chain, bydst) {
2102                 err = xfrm_policy_match(pol, fl, type, family, dir, if_id);
2103                 if (err) {
2104                         if (err == -ESRCH)
2105                                 continue;
2106                         else {
2107                                 ret = ERR_PTR(err);
2108                                 goto fail;
2109                         }
2110                 } else {
2111                         ret = pol;
2112                         break;
2113                 }
2114         }
2115         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2116         if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2117                                                          daddr))
2118                 goto skip_inexact;
2119
2120         pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2121                                           family, dir, if_id);
2122         if (pol) {
2123                 ret = pol;
2124                 if (IS_ERR(pol))
2125                         goto fail;
2126         }
2127
2128 skip_inexact:
2129         if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2130                 goto retry;
2131
2132         if (ret && !xfrm_pol_hold_rcu(ret))
2133                 goto retry;
2134 fail:
2135         rcu_read_unlock();
2136
2137         return ret;
2138 }
2139
2140 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2141                                               const struct flowi *fl,
2142                                               u16 family, u8 dir, u32 if_id)
2143 {
2144 #ifdef CONFIG_XFRM_SUB_POLICY
2145         struct xfrm_policy *pol;
2146
2147         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2148                                         dir, if_id);
2149         if (pol != NULL)
2150                 return pol;
2151 #endif
2152         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2153                                          dir, if_id);
2154 }
2155
2156 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2157                                                  const struct flowi *fl,
2158                                                  u16 family, u32 if_id)
2159 {
2160         struct xfrm_policy *pol;
2161
2162         rcu_read_lock();
2163  again:
2164         pol = rcu_dereference(sk->sk_policy[dir]);
2165         if (pol != NULL) {
2166                 bool match;
2167                 int err = 0;
2168
2169                 if (pol->family != family) {
2170                         pol = NULL;
2171                         goto out;
2172                 }
2173
2174                 match = xfrm_selector_match(&pol->selector, fl, family);
2175                 if (match) {
2176                         if ((sk->sk_mark & pol->mark.m) != pol->mark.v ||
2177                             pol->if_id != if_id) {
2178                                 pol = NULL;
2179                                 goto out;
2180                         }
2181                         err = security_xfrm_policy_lookup(pol->security,
2182                                                       fl->flowi_secid);
2183                         if (!err) {
2184                                 if (!xfrm_pol_hold_rcu(pol))
2185                                         goto again;
2186                         } else if (err == -ESRCH) {
2187                                 pol = NULL;
2188                         } else {
2189                                 pol = ERR_PTR(err);
2190                         }
2191                 } else
2192                         pol = NULL;
2193         }
2194 out:
2195         rcu_read_unlock();
2196         return pol;
2197 }
2198
2199 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2200 {
2201         struct net *net = xp_net(pol);
2202
2203         list_add(&pol->walk.all, &net->xfrm.policy_all);
2204         net->xfrm.policy_count[dir]++;
2205         xfrm_pol_hold(pol);
2206 }
2207
2208 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2209                                                 int dir)
2210 {
2211         struct net *net = xp_net(pol);
2212
2213         if (list_empty(&pol->walk.all))
2214                 return NULL;
2215
2216         /* Socket policies are not hashed. */
2217         if (!hlist_unhashed(&pol->bydst)) {
2218                 hlist_del_rcu(&pol->bydst);
2219                 hlist_del_init(&pol->bydst_inexact_list);
2220                 hlist_del(&pol->byidx);
2221         }
2222
2223         list_del_init(&pol->walk.all);
2224         net->xfrm.policy_count[dir]--;
2225
2226         return pol;
2227 }
2228
2229 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2230 {
2231         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2232 }
2233
2234 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2235 {
2236         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2237 }
2238
2239 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2240 {
2241         struct net *net = xp_net(pol);
2242
2243         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2244         pol = __xfrm_policy_unlink(pol, dir);
2245         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2246         if (pol) {
2247                 xfrm_policy_kill(pol);
2248                 return 0;
2249         }
2250         return -ENOENT;
2251 }
2252 EXPORT_SYMBOL(xfrm_policy_delete);
2253
2254 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2255 {
2256         struct net *net = sock_net(sk);
2257         struct xfrm_policy *old_pol;
2258
2259 #ifdef CONFIG_XFRM_SUB_POLICY
2260         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2261                 return -EINVAL;
2262 #endif
2263
2264         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2265         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2266                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2267         if (pol) {
2268                 pol->curlft.add_time = ktime_get_real_seconds();
2269                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2270                 xfrm_sk_policy_link(pol, dir);
2271         }
2272         rcu_assign_pointer(sk->sk_policy[dir], pol);
2273         if (old_pol) {
2274                 if (pol)
2275                         xfrm_policy_requeue(old_pol, pol);
2276
2277                 /* Unlinking succeeds always. This is the only function
2278                  * allowed to delete or replace socket policy.
2279                  */
2280                 xfrm_sk_policy_unlink(old_pol, dir);
2281         }
2282         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2283
2284         if (old_pol) {
2285                 xfrm_policy_kill(old_pol);
2286         }
2287         return 0;
2288 }
2289
2290 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2291 {
2292         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2293         struct net *net = xp_net(old);
2294
2295         if (newp) {
2296                 newp->selector = old->selector;
2297                 if (security_xfrm_policy_clone(old->security,
2298                                                &newp->security)) {
2299                         kfree(newp);
2300                         return NULL;  /* ENOMEM */
2301                 }
2302                 newp->lft = old->lft;
2303                 newp->curlft = old->curlft;
2304                 newp->mark = old->mark;
2305                 newp->if_id = old->if_id;
2306                 newp->action = old->action;
2307                 newp->flags = old->flags;
2308                 newp->xfrm_nr = old->xfrm_nr;
2309                 newp->index = old->index;
2310                 newp->type = old->type;
2311                 newp->family = old->family;
2312                 memcpy(newp->xfrm_vec, old->xfrm_vec,
2313                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2314                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2315                 xfrm_sk_policy_link(newp, dir);
2316                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2317                 xfrm_pol_put(newp);
2318         }
2319         return newp;
2320 }
2321
2322 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2323 {
2324         const struct xfrm_policy *p;
2325         struct xfrm_policy *np;
2326         int i, ret = 0;
2327
2328         rcu_read_lock();
2329         for (i = 0; i < 2; i++) {
2330                 p = rcu_dereference(osk->sk_policy[i]);
2331                 if (p) {
2332                         np = clone_policy(p, i);
2333                         if (unlikely(!np)) {
2334                                 ret = -ENOMEM;
2335                                 break;
2336                         }
2337                         rcu_assign_pointer(sk->sk_policy[i], np);
2338                 }
2339         }
2340         rcu_read_unlock();
2341         return ret;
2342 }
2343
2344 static int
2345 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local,
2346                xfrm_address_t *remote, unsigned short family, u32 mark)
2347 {
2348         int err;
2349         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2350
2351         if (unlikely(afinfo == NULL))
2352                 return -EINVAL;
2353         err = afinfo->get_saddr(net, oif, local, remote, mark);
2354         rcu_read_unlock();
2355         return err;
2356 }
2357
2358 /* Resolve list of templates for the flow, given policy. */
2359
2360 static int
2361 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2362                       struct xfrm_state **xfrm, unsigned short family)
2363 {
2364         struct net *net = xp_net(policy);
2365         int nx;
2366         int i, error;
2367         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2368         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2369         xfrm_address_t tmp;
2370
2371         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2372                 struct xfrm_state *x;
2373                 xfrm_address_t *remote = daddr;
2374                 xfrm_address_t *local  = saddr;
2375                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2376
2377                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2378                     tmpl->mode == XFRM_MODE_BEET) {
2379                         remote = &tmpl->id.daddr;
2380                         local = &tmpl->saddr;
2381                         if (xfrm_addr_any(local, tmpl->encap_family)) {
2382                                 error = xfrm_get_saddr(net, fl->flowi_oif,
2383                                                        &tmp, remote,
2384                                                        tmpl->encap_family, 0);
2385                                 if (error)
2386                                         goto fail;
2387                                 local = &tmp;
2388                         }
2389                 }
2390
2391                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2392                                     family, policy->if_id);
2393
2394                 if (x && x->km.state == XFRM_STATE_VALID) {
2395                         xfrm[nx++] = x;
2396                         daddr = remote;
2397                         saddr = local;
2398                         continue;
2399                 }
2400                 if (x) {
2401                         error = (x->km.state == XFRM_STATE_ERROR ?
2402                                  -EINVAL : -EAGAIN);
2403                         xfrm_state_put(x);
2404                 } else if (error == -ESRCH) {
2405                         error = -EAGAIN;
2406                 }
2407
2408                 if (!tmpl->optional)
2409                         goto fail;
2410         }
2411         return nx;
2412
2413 fail:
2414         for (nx--; nx >= 0; nx--)
2415                 xfrm_state_put(xfrm[nx]);
2416         return error;
2417 }
2418
2419 static int
2420 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2421                   struct xfrm_state **xfrm, unsigned short family)
2422 {
2423         struct xfrm_state *tp[XFRM_MAX_DEPTH];
2424         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2425         int cnx = 0;
2426         int error;
2427         int ret;
2428         int i;
2429
2430         for (i = 0; i < npols; i++) {
2431                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2432                         error = -ENOBUFS;
2433                         goto fail;
2434                 }
2435
2436                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2437                 if (ret < 0) {
2438                         error = ret;
2439                         goto fail;
2440                 } else
2441                         cnx += ret;
2442         }
2443
2444         /* found states are sorted for outbound processing */
2445         if (npols > 1)
2446                 xfrm_state_sort(xfrm, tpp, cnx, family);
2447
2448         return cnx;
2449
2450  fail:
2451         for (cnx--; cnx >= 0; cnx--)
2452                 xfrm_state_put(tpp[cnx]);
2453         return error;
2454
2455 }
2456
2457 static int xfrm_get_tos(const struct flowi *fl, int family)
2458 {
2459         if (family == AF_INET)
2460                 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos;
2461
2462         return 0;
2463 }
2464
2465 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2466 {
2467         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2468         struct dst_ops *dst_ops;
2469         struct xfrm_dst *xdst;
2470
2471         if (!afinfo)
2472                 return ERR_PTR(-EINVAL);
2473
2474         switch (family) {
2475         case AF_INET:
2476                 dst_ops = &net->xfrm.xfrm4_dst_ops;
2477                 break;
2478 #if IS_ENABLED(CONFIG_IPV6)
2479         case AF_INET6:
2480                 dst_ops = &net->xfrm.xfrm6_dst_ops;
2481                 break;
2482 #endif
2483         default:
2484                 BUG();
2485         }
2486         xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2487
2488         if (likely(xdst)) {
2489                 struct dst_entry *dst = &xdst->u.dst;
2490
2491                 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst));
2492         } else
2493                 xdst = ERR_PTR(-ENOBUFS);
2494
2495         rcu_read_unlock();
2496
2497         return xdst;
2498 }
2499
2500 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2501                            int nfheader_len)
2502 {
2503         if (dst->ops->family == AF_INET6) {
2504                 struct rt6_info *rt = (struct rt6_info *)dst;
2505                 path->path_cookie = rt6_get_cookie(rt);
2506                 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2507         }
2508 }
2509
2510 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2511                                 const struct flowi *fl)
2512 {
2513         const struct xfrm_policy_afinfo *afinfo =
2514                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2515         int err;
2516
2517         if (!afinfo)
2518                 return -EINVAL;
2519
2520         err = afinfo->fill_dst(xdst, dev, fl);
2521
2522         rcu_read_unlock();
2523
2524         return err;
2525 }
2526
2527
2528 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2529  * all the metrics... Shortly, bundle a bundle.
2530  */
2531
2532 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2533                                             struct xfrm_state **xfrm,
2534                                             struct xfrm_dst **bundle,
2535                                             int nx,
2536                                             const struct flowi *fl,
2537                                             struct dst_entry *dst)
2538 {
2539         const struct xfrm_state_afinfo *afinfo;
2540         const struct xfrm_mode *inner_mode;
2541         struct net *net = xp_net(policy);
2542         unsigned long now = jiffies;
2543         struct net_device *dev;
2544         struct xfrm_dst *xdst_prev = NULL;
2545         struct xfrm_dst *xdst0 = NULL;
2546         int i = 0;
2547         int err;
2548         int header_len = 0;
2549         int nfheader_len = 0;
2550         int trailer_len = 0;
2551         int tos;
2552         int family = policy->selector.family;
2553         xfrm_address_t saddr, daddr;
2554
2555         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2556
2557         tos = xfrm_get_tos(fl, family);
2558
2559         dst_hold(dst);
2560
2561         for (; i < nx; i++) {
2562                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2563                 struct dst_entry *dst1 = &xdst->u.dst;
2564
2565                 err = PTR_ERR(xdst);
2566                 if (IS_ERR(xdst)) {
2567                         dst_release(dst);
2568                         goto put_states;
2569                 }
2570
2571                 bundle[i] = xdst;
2572                 if (!xdst_prev)
2573                         xdst0 = xdst;
2574                 else
2575                         /* Ref count is taken during xfrm_alloc_dst()
2576                          * No need to do dst_clone() on dst1
2577                          */
2578                         xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2579
2580                 if (xfrm[i]->sel.family == AF_UNSPEC) {
2581                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
2582                                                         xfrm_af2proto(family));
2583                         if (!inner_mode) {
2584                                 err = -EAFNOSUPPORT;
2585                                 dst_release(dst);
2586                                 goto put_states;
2587                         }
2588                 } else
2589                         inner_mode = &xfrm[i]->inner_mode;
2590
2591                 xdst->route = dst;
2592                 dst_copy_metrics(dst1, dst);
2593
2594                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2595                         __u32 mark = 0;
2596
2597                         if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2598                                 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2599
2600                         family = xfrm[i]->props.family;
2601                         dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif,
2602                                               &saddr, &daddr, family, mark);
2603                         err = PTR_ERR(dst);
2604                         if (IS_ERR(dst))
2605                                 goto put_states;
2606                 } else
2607                         dst_hold(dst);
2608
2609                 dst1->xfrm = xfrm[i];
2610                 xdst->xfrm_genid = xfrm[i]->genid;
2611
2612                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2613                 dst1->lastuse = now;
2614
2615                 dst1->input = dst_discard;
2616
2617                 rcu_read_lock();
2618                 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2619                 if (likely(afinfo))
2620                         dst1->output = afinfo->output;
2621                 else
2622                         dst1->output = dst_discard_out;
2623                 rcu_read_unlock();
2624
2625                 xdst_prev = xdst;
2626
2627                 header_len += xfrm[i]->props.header_len;
2628                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2629                         nfheader_len += xfrm[i]->props.header_len;
2630                 trailer_len += xfrm[i]->props.trailer_len;
2631         }
2632
2633         xfrm_dst_set_child(xdst_prev, dst);
2634         xdst0->path = dst;
2635
2636         err = -ENODEV;
2637         dev = dst->dev;
2638         if (!dev)
2639                 goto free_dst;
2640
2641         xfrm_init_path(xdst0, dst, nfheader_len);
2642         xfrm_init_pmtu(bundle, nx);
2643
2644         for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2645              xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2646                 err = xfrm_fill_dst(xdst_prev, dev, fl);
2647                 if (err)
2648                         goto free_dst;
2649
2650                 xdst_prev->u.dst.header_len = header_len;
2651                 xdst_prev->u.dst.trailer_len = trailer_len;
2652                 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2653                 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2654         }
2655
2656         return &xdst0->u.dst;
2657
2658 put_states:
2659         for (; i < nx; i++)
2660                 xfrm_state_put(xfrm[i]);
2661 free_dst:
2662         if (xdst0)
2663                 dst_release_immediate(&xdst0->u.dst);
2664
2665         return ERR_PTR(err);
2666 }
2667
2668 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2669                                 struct xfrm_policy **pols,
2670                                 int *num_pols, int *num_xfrms)
2671 {
2672         int i;
2673
2674         if (*num_pols == 0 || !pols[0]) {
2675                 *num_pols = 0;
2676                 *num_xfrms = 0;
2677                 return 0;
2678         }
2679         if (IS_ERR(pols[0]))
2680                 return PTR_ERR(pols[0]);
2681
2682         *num_xfrms = pols[0]->xfrm_nr;
2683
2684 #ifdef CONFIG_XFRM_SUB_POLICY
2685         if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW &&
2686             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2687                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2688                                                     XFRM_POLICY_TYPE_MAIN,
2689                                                     fl, family,
2690                                                     XFRM_POLICY_OUT,
2691                                                     pols[0]->if_id);
2692                 if (pols[1]) {
2693                         if (IS_ERR(pols[1])) {
2694                                 xfrm_pols_put(pols, *num_pols);
2695                                 return PTR_ERR(pols[1]);
2696                         }
2697                         (*num_pols)++;
2698                         (*num_xfrms) += pols[1]->xfrm_nr;
2699                 }
2700         }
2701 #endif
2702         for (i = 0; i < *num_pols; i++) {
2703                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2704                         *num_xfrms = -1;
2705                         break;
2706                 }
2707         }
2708
2709         return 0;
2710
2711 }
2712
2713 static struct xfrm_dst *
2714 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2715                                const struct flowi *fl, u16 family,
2716                                struct dst_entry *dst_orig)
2717 {
2718         struct net *net = xp_net(pols[0]);
2719         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2720         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2721         struct xfrm_dst *xdst;
2722         struct dst_entry *dst;
2723         int err;
2724
2725         /* Try to instantiate a bundle */
2726         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2727         if (err <= 0) {
2728                 if (err == 0)
2729                         return NULL;
2730
2731                 if (err != -EAGAIN)
2732                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2733                 return ERR_PTR(err);
2734         }
2735
2736         dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2737         if (IS_ERR(dst)) {
2738                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2739                 return ERR_CAST(dst);
2740         }
2741
2742         xdst = (struct xfrm_dst *)dst;
2743         xdst->num_xfrms = err;
2744         xdst->num_pols = num_pols;
2745         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2746         xdst->policy_genid = atomic_read(&pols[0]->genid);
2747
2748         return xdst;
2749 }
2750
2751 static void xfrm_policy_queue_process(struct timer_list *t)
2752 {
2753         struct sk_buff *skb;
2754         struct sock *sk;
2755         struct dst_entry *dst;
2756         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2757         struct net *net = xp_net(pol);
2758         struct xfrm_policy_queue *pq = &pol->polq;
2759         struct flowi fl;
2760         struct sk_buff_head list;
2761         __u32 skb_mark;
2762
2763         spin_lock(&pq->hold_queue.lock);
2764         skb = skb_peek(&pq->hold_queue);
2765         if (!skb) {
2766                 spin_unlock(&pq->hold_queue.lock);
2767                 goto out;
2768         }
2769         dst = skb_dst(skb);
2770         sk = skb->sk;
2771
2772         /* Fixup the mark to support VTI. */
2773         skb_mark = skb->mark;
2774         skb->mark = pol->mark.v;
2775         xfrm_decode_session(skb, &fl, dst->ops->family);
2776         skb->mark = skb_mark;
2777         spin_unlock(&pq->hold_queue.lock);
2778
2779         dst_hold(xfrm_dst_path(dst));
2780         dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2781         if (IS_ERR(dst))
2782                 goto purge_queue;
2783
2784         if (dst->flags & DST_XFRM_QUEUE) {
2785                 dst_release(dst);
2786
2787                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2788                         goto purge_queue;
2789
2790                 pq->timeout = pq->timeout << 1;
2791                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2792                         xfrm_pol_hold(pol);
2793                 goto out;
2794         }
2795
2796         dst_release(dst);
2797
2798         __skb_queue_head_init(&list);
2799
2800         spin_lock(&pq->hold_queue.lock);
2801         pq->timeout = 0;
2802         skb_queue_splice_init(&pq->hold_queue, &list);
2803         spin_unlock(&pq->hold_queue.lock);
2804
2805         while (!skb_queue_empty(&list)) {
2806                 skb = __skb_dequeue(&list);
2807
2808                 /* Fixup the mark to support VTI. */
2809                 skb_mark = skb->mark;
2810                 skb->mark = pol->mark.v;
2811                 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family);
2812                 skb->mark = skb_mark;
2813
2814                 dst_hold(xfrm_dst_path(skb_dst(skb)));
2815                 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2816                 if (IS_ERR(dst)) {
2817                         kfree_skb(skb);
2818                         continue;
2819                 }
2820
2821                 nf_reset_ct(skb);
2822                 skb_dst_drop(skb);
2823                 skb_dst_set(skb, dst);
2824
2825                 dst_output(net, skb->sk, skb);
2826         }
2827
2828 out:
2829         xfrm_pol_put(pol);
2830         return;
2831
2832 purge_queue:
2833         pq->timeout = 0;
2834         skb_queue_purge(&pq->hold_queue);
2835         xfrm_pol_put(pol);
2836 }
2837
2838 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2839 {
2840         unsigned long sched_next;
2841         struct dst_entry *dst = skb_dst(skb);
2842         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2843         struct xfrm_policy *pol = xdst->pols[0];
2844         struct xfrm_policy_queue *pq = &pol->polq;
2845
2846         if (unlikely(skb_fclone_busy(sk, skb))) {
2847                 kfree_skb(skb);
2848                 return 0;
2849         }
2850
2851         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2852                 kfree_skb(skb);
2853                 return -EAGAIN;
2854         }
2855
2856         skb_dst_force(skb);
2857
2858         spin_lock_bh(&pq->hold_queue.lock);
2859
2860         if (!pq->timeout)
2861                 pq->timeout = XFRM_QUEUE_TMO_MIN;
2862
2863         sched_next = jiffies + pq->timeout;
2864
2865         if (del_timer(&pq->hold_timer)) {
2866                 if (time_before(pq->hold_timer.expires, sched_next))
2867                         sched_next = pq->hold_timer.expires;
2868                 xfrm_pol_put(pol);
2869         }
2870
2871         __skb_queue_tail(&pq->hold_queue, skb);
2872         if (!mod_timer(&pq->hold_timer, sched_next))
2873                 xfrm_pol_hold(pol);
2874
2875         spin_unlock_bh(&pq->hold_queue.lock);
2876
2877         return 0;
2878 }
2879
2880 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
2881                                                  struct xfrm_flo *xflo,
2882                                                  const struct flowi *fl,
2883                                                  int num_xfrms,
2884                                                  u16 family)
2885 {
2886         int err;
2887         struct net_device *dev;
2888         struct dst_entry *dst;
2889         struct dst_entry *dst1;
2890         struct xfrm_dst *xdst;
2891
2892         xdst = xfrm_alloc_dst(net, family);
2893         if (IS_ERR(xdst))
2894                 return xdst;
2895
2896         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
2897             net->xfrm.sysctl_larval_drop ||
2898             num_xfrms <= 0)
2899                 return xdst;
2900
2901         dst = xflo->dst_orig;
2902         dst1 = &xdst->u.dst;
2903         dst_hold(dst);
2904         xdst->route = dst;
2905
2906         dst_copy_metrics(dst1, dst);
2907
2908         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2909         dst1->flags |= DST_XFRM_QUEUE;
2910         dst1->lastuse = jiffies;
2911
2912         dst1->input = dst_discard;
2913         dst1->output = xdst_queue_output;
2914
2915         dst_hold(dst);
2916         xfrm_dst_set_child(xdst, dst);
2917         xdst->path = dst;
2918
2919         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
2920
2921         err = -ENODEV;
2922         dev = dst->dev;
2923         if (!dev)
2924                 goto free_dst;
2925
2926         err = xfrm_fill_dst(xdst, dev, fl);
2927         if (err)
2928                 goto free_dst;
2929
2930 out:
2931         return xdst;
2932
2933 free_dst:
2934         dst_release(dst1);
2935         xdst = ERR_PTR(err);
2936         goto out;
2937 }
2938
2939 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
2940                                            const struct flowi *fl,
2941                                            u16 family, u8 dir,
2942                                            struct xfrm_flo *xflo, u32 if_id)
2943 {
2944         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
2945         int num_pols = 0, num_xfrms = 0, err;
2946         struct xfrm_dst *xdst;
2947
2948         /* Resolve policies to use if we couldn't get them from
2949          * previous cache entry */
2950         num_pols = 1;
2951         pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
2952         err = xfrm_expand_policies(fl, family, pols,
2953                                            &num_pols, &num_xfrms);
2954         if (err < 0)
2955                 goto inc_error;
2956         if (num_pols == 0)
2957                 return NULL;
2958         if (num_xfrms <= 0)
2959                 goto make_dummy_bundle;
2960
2961         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
2962                                               xflo->dst_orig);
2963         if (IS_ERR(xdst)) {
2964                 err = PTR_ERR(xdst);
2965                 if (err == -EREMOTE) {
2966                         xfrm_pols_put(pols, num_pols);
2967                         return NULL;
2968                 }
2969
2970                 if (err != -EAGAIN)
2971                         goto error;
2972                 goto make_dummy_bundle;
2973         } else if (xdst == NULL) {
2974                 num_xfrms = 0;
2975                 goto make_dummy_bundle;
2976         }
2977
2978         return xdst;
2979
2980 make_dummy_bundle:
2981         /* We found policies, but there's no bundles to instantiate:
2982          * either because the policy blocks, has no transformations or
2983          * we could not build template (no xfrm_states).*/
2984         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
2985         if (IS_ERR(xdst)) {
2986                 xfrm_pols_put(pols, num_pols);
2987                 return ERR_CAST(xdst);
2988         }
2989         xdst->num_pols = num_pols;
2990         xdst->num_xfrms = num_xfrms;
2991         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2992
2993         return xdst;
2994
2995 inc_error:
2996         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2997 error:
2998         xfrm_pols_put(pols, num_pols);
2999         return ERR_PTR(err);
3000 }
3001
3002 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3003                                         struct dst_entry *dst_orig)
3004 {
3005         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3006         struct dst_entry *ret;
3007
3008         if (!afinfo) {
3009                 dst_release(dst_orig);
3010                 return ERR_PTR(-EINVAL);
3011         } else {
3012                 ret = afinfo->blackhole_route(net, dst_orig);
3013         }
3014         rcu_read_unlock();
3015
3016         return ret;
3017 }
3018
3019 /* Finds/creates a bundle for given flow and if_id
3020  *
3021  * At the moment we eat a raw IP route. Mostly to speed up lookups
3022  * on interfaces with disabled IPsec.
3023  *
3024  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3025  * compatibility
3026  */
3027 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3028                                         struct dst_entry *dst_orig,
3029                                         const struct flowi *fl,
3030                                         const struct sock *sk,
3031                                         int flags, u32 if_id)
3032 {
3033         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3034         struct xfrm_dst *xdst;
3035         struct dst_entry *dst, *route;
3036         u16 family = dst_orig->ops->family;
3037         u8 dir = XFRM_POLICY_OUT;
3038         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3039
3040         dst = NULL;
3041         xdst = NULL;
3042         route = NULL;
3043
3044         sk = sk_const_to_full_sk(sk);
3045         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3046                 num_pols = 1;
3047                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3048                                                 if_id);
3049                 err = xfrm_expand_policies(fl, family, pols,
3050                                            &num_pols, &num_xfrms);
3051                 if (err < 0)
3052                         goto dropdst;
3053
3054                 if (num_pols) {
3055                         if (num_xfrms <= 0) {
3056                                 drop_pols = num_pols;
3057                                 goto no_transform;
3058                         }
3059
3060                         xdst = xfrm_resolve_and_create_bundle(
3061                                         pols, num_pols, fl,
3062                                         family, dst_orig);
3063
3064                         if (IS_ERR(xdst)) {
3065                                 xfrm_pols_put(pols, num_pols);
3066                                 err = PTR_ERR(xdst);
3067                                 if (err == -EREMOTE)
3068                                         goto nopol;
3069
3070                                 goto dropdst;
3071                         } else if (xdst == NULL) {
3072                                 num_xfrms = 0;
3073                                 drop_pols = num_pols;
3074                                 goto no_transform;
3075                         }
3076
3077                         route = xdst->route;
3078                 }
3079         }
3080
3081         if (xdst == NULL) {
3082                 struct xfrm_flo xflo;
3083
3084                 xflo.dst_orig = dst_orig;
3085                 xflo.flags = flags;
3086
3087                 /* To accelerate a bit...  */
3088                 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3089                                !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3090                         goto nopol;
3091
3092                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3093                 if (xdst == NULL)
3094                         goto nopol;
3095                 if (IS_ERR(xdst)) {
3096                         err = PTR_ERR(xdst);
3097                         goto dropdst;
3098                 }
3099
3100                 num_pols = xdst->num_pols;
3101                 num_xfrms = xdst->num_xfrms;
3102                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3103                 route = xdst->route;
3104         }
3105
3106         dst = &xdst->u.dst;
3107         if (route == NULL && num_xfrms > 0) {
3108                 /* The only case when xfrm_bundle_lookup() returns a
3109                  * bundle with null route, is when the template could
3110                  * not be resolved. It means policies are there, but
3111                  * bundle could not be created, since we don't yet
3112                  * have the xfrm_state's. We need to wait for KM to
3113                  * negotiate new SA's or bail out with error.*/
3114                 if (net->xfrm.sysctl_larval_drop) {
3115                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3116                         err = -EREMOTE;
3117                         goto error;
3118                 }
3119
3120                 err = -EAGAIN;
3121
3122                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3123                 goto error;
3124         }
3125
3126 no_transform:
3127         if (num_pols == 0)
3128                 goto nopol;
3129
3130         if ((flags & XFRM_LOOKUP_ICMP) &&
3131             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3132                 err = -ENOENT;
3133                 goto error;
3134         }
3135
3136         for (i = 0; i < num_pols; i++)
3137                 pols[i]->curlft.use_time = ktime_get_real_seconds();
3138
3139         if (num_xfrms < 0) {
3140                 /* Prohibit the flow */
3141                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3142                 err = -EPERM;
3143                 goto error;
3144         } else if (num_xfrms > 0) {
3145                 /* Flow transformed */
3146                 dst_release(dst_orig);
3147         } else {
3148                 /* Flow passes untransformed */
3149                 dst_release(dst);
3150                 dst = dst_orig;
3151         }
3152 ok:
3153         xfrm_pols_put(pols, drop_pols);
3154         if (dst && dst->xfrm &&
3155             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3156                 dst->flags |= DST_XFRM_TUNNEL;
3157         return dst;
3158
3159 nopol:
3160         if (!(dst_orig->dev->flags & IFF_LOOPBACK) &&
3161             !xfrm_default_allow(net, dir)) {
3162                 err = -EPERM;
3163                 goto error;
3164         }
3165         if (!(flags & XFRM_LOOKUP_ICMP)) {
3166                 dst = dst_orig;
3167                 goto ok;
3168         }
3169         err = -ENOENT;
3170 error:
3171         dst_release(dst);
3172 dropdst:
3173         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3174                 dst_release(dst_orig);
3175         xfrm_pols_put(pols, drop_pols);
3176         return ERR_PTR(err);
3177 }
3178 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3179
3180 /* Main function: finds/creates a bundle for given flow.
3181  *
3182  * At the moment we eat a raw IP route. Mostly to speed up lookups
3183  * on interfaces with disabled IPsec.
3184  */
3185 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3186                               const struct flowi *fl, const struct sock *sk,
3187                               int flags)
3188 {
3189         return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3190 }
3191 EXPORT_SYMBOL(xfrm_lookup);
3192
3193 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3194  * Otherwise we may send out blackholed packets.
3195  */
3196 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3197                                     const struct flowi *fl,
3198                                     const struct sock *sk, int flags)
3199 {
3200         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3201                                             flags | XFRM_LOOKUP_QUEUE |
3202                                             XFRM_LOOKUP_KEEP_DST_REF);
3203
3204         if (PTR_ERR(dst) == -EREMOTE)
3205                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3206
3207         if (IS_ERR(dst))
3208                 dst_release(dst_orig);
3209
3210         return dst;
3211 }
3212 EXPORT_SYMBOL(xfrm_lookup_route);
3213
3214 static inline int
3215 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3216 {
3217         struct sec_path *sp = skb_sec_path(skb);
3218         struct xfrm_state *x;
3219
3220         if (!sp || idx < 0 || idx >= sp->len)
3221                 return 0;
3222         x = sp->xvec[idx];
3223         if (!x->type->reject)
3224                 return 0;
3225         return x->type->reject(x, skb, fl);
3226 }
3227
3228 /* When skb is transformed back to its "native" form, we have to
3229  * check policy restrictions. At the moment we make this in maximally
3230  * stupid way. Shame on me. :-) Of course, connected sockets must
3231  * have policy cached at them.
3232  */
3233
3234 static inline int
3235 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3236               unsigned short family)
3237 {
3238         if (xfrm_state_kern(x))
3239                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3240         return  x->id.proto == tmpl->id.proto &&
3241                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3242                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3243                 x->props.mode == tmpl->mode &&
3244                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3245                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3246                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3247                   xfrm_state_addr_cmp(tmpl, x, family));
3248 }
3249
3250 /*
3251  * 0 or more than 0 is returned when validation is succeeded (either bypass
3252  * because of optional transport mode, or next index of the matched secpath
3253  * state with the template.
3254  * -1 is returned when no matching template is found.
3255  * Otherwise "-2 - errored_index" is returned.
3256  */
3257 static inline int
3258 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3259                unsigned short family)
3260 {
3261         int idx = start;
3262
3263         if (tmpl->optional) {
3264                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3265                         return start;
3266         } else
3267                 start = -1;
3268         for (; idx < sp->len; idx++) {
3269                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
3270                         return ++idx;
3271                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3272                         if (start == -1)
3273                                 start = -2-idx;
3274                         break;
3275                 }
3276         }
3277         return start;
3278 }
3279
3280 static void
3281 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse)
3282 {
3283         const struct iphdr *iph = ip_hdr(skb);
3284         int ihl = iph->ihl;
3285         u8 *xprth = skb_network_header(skb) + ihl * 4;
3286         struct flowi4 *fl4 = &fl->u.ip4;
3287         int oif = 0;
3288
3289         if (skb_dst(skb) && skb_dst(skb)->dev)
3290                 oif = skb_dst(skb)->dev->ifindex;
3291
3292         memset(fl4, 0, sizeof(struct flowi4));
3293         fl4->flowi4_mark = skb->mark;
3294         fl4->flowi4_oif = reverse ? skb->skb_iif : oif;
3295
3296         fl4->flowi4_proto = iph->protocol;
3297         fl4->daddr = reverse ? iph->saddr : iph->daddr;
3298         fl4->saddr = reverse ? iph->daddr : iph->saddr;
3299         fl4->flowi4_tos = iph->tos;
3300
3301         if (!ip_is_fragment(iph)) {
3302                 switch (iph->protocol) {
3303                 case IPPROTO_UDP:
3304                 case IPPROTO_UDPLITE:
3305                 case IPPROTO_TCP:
3306                 case IPPROTO_SCTP:
3307                 case IPPROTO_DCCP:
3308                         if (xprth + 4 < skb->data ||
3309                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
3310                                 __be16 *ports;
3311
3312                                 xprth = skb_network_header(skb) + ihl * 4;
3313                                 ports = (__be16 *)xprth;
3314
3315                                 fl4->fl4_sport = ports[!!reverse];
3316                                 fl4->fl4_dport = ports[!reverse];
3317                         }
3318                         break;
3319                 case IPPROTO_ICMP:
3320                         if (xprth + 2 < skb->data ||
3321                             pskb_may_pull(skb, xprth + 2 - skb->data)) {
3322                                 u8 *icmp;
3323
3324                                 xprth = skb_network_header(skb) + ihl * 4;
3325                                 icmp = xprth;
3326
3327                                 fl4->fl4_icmp_type = icmp[0];
3328                                 fl4->fl4_icmp_code = icmp[1];
3329                         }
3330                         break;
3331                 case IPPROTO_GRE:
3332                         if (xprth + 12 < skb->data ||
3333                             pskb_may_pull(skb, xprth + 12 - skb->data)) {
3334                                 __be16 *greflags;
3335                                 __be32 *gre_hdr;
3336
3337                                 xprth = skb_network_header(skb) + ihl * 4;
3338                                 greflags = (__be16 *)xprth;
3339                                 gre_hdr = (__be32 *)xprth;
3340
3341                                 if (greflags[0] & GRE_KEY) {
3342                                         if (greflags[0] & GRE_CSUM)
3343                                                 gre_hdr++;
3344                                         fl4->fl4_gre_key = gre_hdr[1];
3345                                 }
3346                         }
3347                         break;
3348                 default:
3349                         break;
3350                 }
3351         }
3352 }
3353
3354 #if IS_ENABLED(CONFIG_IPV6)
3355 static void
3356 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse)
3357 {
3358         struct flowi6 *fl6 = &fl->u.ip6;
3359         int onlyproto = 0;
3360         const struct ipv6hdr *hdr = ipv6_hdr(skb);
3361         u32 offset = sizeof(*hdr);
3362         struct ipv6_opt_hdr *exthdr;
3363         const unsigned char *nh = skb_network_header(skb);
3364         u16 nhoff = IP6CB(skb)->nhoff;
3365         int oif = 0;
3366         u8 nexthdr;
3367
3368         if (!nhoff)
3369                 nhoff = offsetof(struct ipv6hdr, nexthdr);
3370
3371         nexthdr = nh[nhoff];
3372
3373         if (skb_dst(skb) && skb_dst(skb)->dev)
3374                 oif = skb_dst(skb)->dev->ifindex;
3375
3376         memset(fl6, 0, sizeof(struct flowi6));
3377         fl6->flowi6_mark = skb->mark;
3378         fl6->flowi6_oif = reverse ? skb->skb_iif : oif;
3379
3380         fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
3381         fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
3382
3383         while (nh + offset + sizeof(*exthdr) < skb->data ||
3384                pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) {
3385                 nh = skb_network_header(skb);
3386                 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3387
3388                 switch (nexthdr) {
3389                 case NEXTHDR_FRAGMENT:
3390                         onlyproto = 1;
3391                         fallthrough;
3392                 case NEXTHDR_ROUTING:
3393                 case NEXTHDR_HOP:
3394                 case NEXTHDR_DEST:
3395                         offset += ipv6_optlen(exthdr);
3396                         nexthdr = exthdr->nexthdr;
3397                         exthdr = (struct ipv6_opt_hdr *)(nh + offset);
3398                         break;
3399                 case IPPROTO_UDP:
3400                 case IPPROTO_UDPLITE:
3401                 case IPPROTO_TCP:
3402                 case IPPROTO_SCTP:
3403                 case IPPROTO_DCCP:
3404                         if (!onlyproto && (nh + offset + 4 < skb->data ||
3405                              pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
3406                                 __be16 *ports;
3407
3408                                 nh = skb_network_header(skb);
3409                                 ports = (__be16 *)(nh + offset);
3410                                 fl6->fl6_sport = ports[!!reverse];
3411                                 fl6->fl6_dport = ports[!reverse];
3412                         }
3413                         fl6->flowi6_proto = nexthdr;
3414                         return;
3415                 case IPPROTO_ICMPV6:
3416                         if (!onlyproto && (nh + offset + 2 < skb->data ||
3417                             pskb_may_pull(skb, nh + offset + 2 - skb->data))) {
3418                                 u8 *icmp;
3419
3420                                 nh = skb_network_header(skb);
3421                                 icmp = (u8 *)(nh + offset);
3422                                 fl6->fl6_icmp_type = icmp[0];
3423                                 fl6->fl6_icmp_code = icmp[1];
3424                         }
3425                         fl6->flowi6_proto = nexthdr;
3426                         return;
3427 #if IS_ENABLED(CONFIG_IPV6_MIP6)
3428                 case IPPROTO_MH:
3429                         offset += ipv6_optlen(exthdr);
3430                         if (!onlyproto && (nh + offset + 3 < skb->data ||
3431                             pskb_may_pull(skb, nh + offset + 3 - skb->data))) {
3432                                 struct ip6_mh *mh;
3433
3434                                 nh = skb_network_header(skb);
3435                                 mh = (struct ip6_mh *)(nh + offset);
3436                                 fl6->fl6_mh_type = mh->ip6mh_type;
3437                         }
3438                         fl6->flowi6_proto = nexthdr;
3439                         return;
3440 #endif
3441                 default:
3442                         fl6->flowi6_proto = nexthdr;
3443                         return;
3444                 }
3445         }
3446 }
3447 #endif
3448
3449 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl,
3450                           unsigned int family, int reverse)
3451 {
3452         switch (family) {
3453         case AF_INET:
3454                 decode_session4(skb, fl, reverse);
3455                 break;
3456 #if IS_ENABLED(CONFIG_IPV6)
3457         case AF_INET6:
3458                 decode_session6(skb, fl, reverse);
3459                 break;
3460 #endif
3461         default:
3462                 return -EAFNOSUPPORT;
3463         }
3464
3465         return security_xfrm_decode_session(skb, &fl->flowi_secid);
3466 }
3467 EXPORT_SYMBOL(__xfrm_decode_session);
3468
3469 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3470 {
3471         for (; k < sp->len; k++) {
3472                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3473                         *idxp = k;
3474                         return 1;
3475                 }
3476         }
3477
3478         return 0;
3479 }
3480
3481 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3482                         unsigned short family)
3483 {
3484         struct net *net = dev_net(skb->dev);
3485         struct xfrm_policy *pol;
3486         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3487         int npols = 0;
3488         int xfrm_nr;
3489         int pi;
3490         int reverse;
3491         struct flowi fl;
3492         int xerr_idx = -1;
3493         const struct xfrm_if_cb *ifcb;
3494         struct sec_path *sp;
3495         struct xfrm_if *xi;
3496         u32 if_id = 0;
3497
3498         rcu_read_lock();
3499         ifcb = xfrm_if_get_cb();
3500
3501         if (ifcb) {
3502                 xi = ifcb->decode_session(skb, family);
3503                 if (xi) {
3504                         if_id = xi->p.if_id;
3505                         net = xi->net;
3506                 }
3507         }
3508         rcu_read_unlock();
3509
3510         reverse = dir & ~XFRM_POLICY_MASK;
3511         dir &= XFRM_POLICY_MASK;
3512
3513         if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
3514                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3515                 return 0;
3516         }
3517
3518         nf_nat_decode_session(skb, &fl, family);
3519
3520         /* First, check used SA against their selectors. */
3521         sp = skb_sec_path(skb);
3522         if (sp) {
3523                 int i;
3524
3525                 for (i = sp->len - 1; i >= 0; i--) {
3526                         struct xfrm_state *x = sp->xvec[i];
3527                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
3528                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3529                                 return 0;
3530                         }
3531                 }
3532         }
3533
3534         pol = NULL;
3535         sk = sk_to_full_sk(sk);
3536         if (sk && sk->sk_policy[dir]) {
3537                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3538                 if (IS_ERR(pol)) {
3539                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3540                         return 0;
3541                 }
3542         }
3543
3544         if (!pol)
3545                 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3546
3547         if (IS_ERR(pol)) {
3548                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3549                 return 0;
3550         }
3551
3552         if (!pol) {
3553                 if (!xfrm_default_allow(net, dir)) {
3554                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3555                         return 0;
3556                 }
3557
3558                 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) {
3559                         xfrm_secpath_reject(xerr_idx, skb, &fl);
3560                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3561                         return 0;
3562                 }
3563                 return 1;
3564         }
3565
3566         pol->curlft.use_time = ktime_get_real_seconds();
3567
3568         pols[0] = pol;
3569         npols++;
3570 #ifdef CONFIG_XFRM_SUB_POLICY
3571         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3572                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3573                                                     &fl, family,
3574                                                     XFRM_POLICY_IN, if_id);
3575                 if (pols[1]) {
3576                         if (IS_ERR(pols[1])) {
3577                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3578                                 return 0;
3579                         }
3580                         pols[1]->curlft.use_time = ktime_get_real_seconds();
3581                         npols++;
3582                 }
3583         }
3584 #endif
3585
3586         if (pol->action == XFRM_POLICY_ALLOW) {
3587                 static struct sec_path dummy;
3588                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3589                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3590                 struct xfrm_tmpl **tpp = tp;
3591                 int ti = 0;
3592                 int i, k;
3593
3594                 sp = skb_sec_path(skb);
3595                 if (!sp)
3596                         sp = &dummy;
3597
3598                 for (pi = 0; pi < npols; pi++) {
3599                         if (pols[pi] != pol &&
3600                             pols[pi]->action != XFRM_POLICY_ALLOW) {
3601                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3602                                 goto reject;
3603                         }
3604                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3605                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3606                                 goto reject_error;
3607                         }
3608                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
3609                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3610                 }
3611                 xfrm_nr = ti;
3612
3613                 if (!xfrm_default_allow(net, dir) && !xfrm_nr) {
3614                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
3615                         goto reject;
3616                 }
3617
3618                 if (npols > 1) {
3619                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3620                         tpp = stp;
3621                 }
3622
3623                 /* For each tunnel xfrm, find the first matching tmpl.
3624                  * For each tmpl before that, find corresponding xfrm.
3625                  * Order is _important_. Later we will implement
3626                  * some barriers, but at the moment barriers
3627                  * are implied between each two transformations.
3628                  */
3629                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3630                         k = xfrm_policy_ok(tpp[i], sp, k, family);
3631                         if (k < 0) {
3632                                 if (k < -1)
3633                                         /* "-2 - errored_index" returned */
3634                                         xerr_idx = -(2+k);
3635                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3636                                 goto reject;
3637                         }
3638                 }
3639
3640                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3641                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3642                         goto reject;
3643                 }
3644
3645                 xfrm_pols_put(pols, npols);
3646                 return 1;
3647         }
3648         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3649
3650 reject:
3651         xfrm_secpath_reject(xerr_idx, skb, &fl);
3652 reject_error:
3653         xfrm_pols_put(pols, npols);
3654         return 0;
3655 }
3656 EXPORT_SYMBOL(__xfrm_policy_check);
3657
3658 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3659 {
3660         struct net *net = dev_net(skb->dev);
3661         struct flowi fl;
3662         struct dst_entry *dst;
3663         int res = 1;
3664
3665         if (xfrm_decode_session(skb, &fl, family) < 0) {
3666                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3667                 return 0;
3668         }
3669
3670         skb_dst_force(skb);
3671         if (!skb_dst(skb)) {
3672                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3673                 return 0;
3674         }
3675
3676         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3677         if (IS_ERR(dst)) {
3678                 res = 0;
3679                 dst = NULL;
3680         }
3681         skb_dst_set(skb, dst);
3682         return res;
3683 }
3684 EXPORT_SYMBOL(__xfrm_route_forward);
3685
3686 /* Optimize later using cookies and generation ids. */
3687
3688 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3689 {
3690         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3691          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3692          * get validated by dst_ops->check on every use.  We do this
3693          * because when a normal route referenced by an XFRM dst is
3694          * obsoleted we do not go looking around for all parent
3695          * referencing XFRM dsts so that we can invalidate them.  It
3696          * is just too much work.  Instead we make the checks here on
3697          * every use.  For example:
3698          *
3699          *      XFRM dst A --> IPv4 dst X
3700          *
3701          * X is the "xdst->route" of A (X is also the "dst->path" of A
3702          * in this example).  If X is marked obsolete, "A" will not
3703          * notice.  That's what we are validating here via the
3704          * stale_bundle() check.
3705          *
3706          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3707          * be marked on it.
3708          * This will force stale_bundle() to fail on any xdst bundle with
3709          * this dst linked in it.
3710          */
3711         if (dst->obsolete < 0 && !stale_bundle(dst))
3712                 return dst;
3713
3714         return NULL;
3715 }
3716
3717 static int stale_bundle(struct dst_entry *dst)
3718 {
3719         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3720 }
3721
3722 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3723 {
3724         while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3725                 dst->dev = dev_net(dev)->loopback_dev;
3726                 dev_hold(dst->dev);
3727                 dev_put(dev);
3728         }
3729 }
3730 EXPORT_SYMBOL(xfrm_dst_ifdown);
3731
3732 static void xfrm_link_failure(struct sk_buff *skb)
3733 {
3734         /* Impossible. Such dst must be popped before reaches point of failure. */
3735 }
3736
3737 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
3738 {
3739         if (dst) {
3740                 if (dst->obsolete) {
3741                         dst_release(dst);
3742                         dst = NULL;
3743                 }
3744         }
3745         return dst;
3746 }
3747
3748 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3749 {
3750         while (nr--) {
3751                 struct xfrm_dst *xdst = bundle[nr];
3752                 u32 pmtu, route_mtu_cached;
3753                 struct dst_entry *dst;
3754
3755                 dst = &xdst->u.dst;
3756                 pmtu = dst_mtu(xfrm_dst_child(dst));
3757                 xdst->child_mtu_cached = pmtu;
3758
3759                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3760
3761                 route_mtu_cached = dst_mtu(xdst->route);
3762                 xdst->route_mtu_cached = route_mtu_cached;
3763
3764                 if (pmtu > route_mtu_cached)
3765                         pmtu = route_mtu_cached;
3766
3767                 dst_metric_set(dst, RTAX_MTU, pmtu);
3768         }
3769 }
3770
3771 /* Check that the bundle accepts the flow and its components are
3772  * still valid.
3773  */
3774
3775 static int xfrm_bundle_ok(struct xfrm_dst *first)
3776 {
3777         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3778         struct dst_entry *dst = &first->u.dst;
3779         struct xfrm_dst *xdst;
3780         int start_from, nr;
3781         u32 mtu;
3782
3783         if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3784             (dst->dev && !netif_running(dst->dev)))
3785                 return 0;
3786
3787         if (dst->flags & DST_XFRM_QUEUE)
3788                 return 1;
3789
3790         start_from = nr = 0;
3791         do {
3792                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3793
3794                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3795                         return 0;
3796                 if (xdst->xfrm_genid != dst->xfrm->genid)
3797                         return 0;
3798                 if (xdst->num_pols > 0 &&
3799                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
3800                         return 0;
3801
3802                 bundle[nr++] = xdst;
3803
3804                 mtu = dst_mtu(xfrm_dst_child(dst));
3805                 if (xdst->child_mtu_cached != mtu) {
3806                         start_from = nr;
3807                         xdst->child_mtu_cached = mtu;
3808                 }
3809
3810                 if (!dst_check(xdst->route, xdst->route_cookie))
3811                         return 0;
3812                 mtu = dst_mtu(xdst->route);
3813                 if (xdst->route_mtu_cached != mtu) {
3814                         start_from = nr;
3815                         xdst->route_mtu_cached = mtu;
3816                 }
3817
3818                 dst = xfrm_dst_child(dst);
3819         } while (dst->xfrm);
3820
3821         if (likely(!start_from))
3822                 return 1;
3823
3824         xdst = bundle[start_from - 1];
3825         mtu = xdst->child_mtu_cached;
3826         while (start_from--) {
3827                 dst = &xdst->u.dst;
3828
3829                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
3830                 if (mtu > xdst->route_mtu_cached)
3831                         mtu = xdst->route_mtu_cached;
3832                 dst_metric_set(dst, RTAX_MTU, mtu);
3833                 if (!start_from)
3834                         break;
3835
3836                 xdst = bundle[start_from - 1];
3837                 xdst->child_mtu_cached = mtu;
3838         }
3839
3840         return 1;
3841 }
3842
3843 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
3844 {
3845         return dst_metric_advmss(xfrm_dst_path(dst));
3846 }
3847
3848 static unsigned int xfrm_mtu(const struct dst_entry *dst)
3849 {
3850         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
3851
3852         return mtu ? : dst_mtu(xfrm_dst_path(dst));
3853 }
3854
3855 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
3856                                         const void *daddr)
3857 {
3858         while (dst->xfrm) {
3859                 const struct xfrm_state *xfrm = dst->xfrm;
3860
3861                 dst = xfrm_dst_child(dst);
3862
3863                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
3864                         continue;
3865                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
3866                         daddr = xfrm->coaddr;
3867                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
3868                         daddr = &xfrm->id.daddr;
3869         }
3870         return daddr;
3871 }
3872
3873 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
3874                                            struct sk_buff *skb,
3875                                            const void *daddr)
3876 {
3877         const struct dst_entry *path = xfrm_dst_path(dst);
3878
3879         if (!skb)
3880                 daddr = xfrm_get_dst_nexthop(dst, daddr);
3881         return path->ops->neigh_lookup(path, skb, daddr);
3882 }
3883
3884 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
3885 {
3886         const struct dst_entry *path = xfrm_dst_path(dst);
3887
3888         daddr = xfrm_get_dst_nexthop(dst, daddr);
3889         path->ops->confirm_neigh(path, daddr);
3890 }
3891
3892 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
3893 {
3894         int err = 0;
3895
3896         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
3897                 return -EAFNOSUPPORT;
3898
3899         spin_lock(&xfrm_policy_afinfo_lock);
3900         if (unlikely(xfrm_policy_afinfo[family] != NULL))
3901                 err = -EEXIST;
3902         else {
3903                 struct dst_ops *dst_ops = afinfo->dst_ops;
3904                 if (likely(dst_ops->kmem_cachep == NULL))
3905                         dst_ops->kmem_cachep = xfrm_dst_cache;
3906                 if (likely(dst_ops->check == NULL))
3907                         dst_ops->check = xfrm_dst_check;
3908                 if (likely(dst_ops->default_advmss == NULL))
3909                         dst_ops->default_advmss = xfrm_default_advmss;
3910                 if (likely(dst_ops->mtu == NULL))
3911                         dst_ops->mtu = xfrm_mtu;
3912                 if (likely(dst_ops->negative_advice == NULL))
3913                         dst_ops->negative_advice = xfrm_negative_advice;
3914                 if (likely(dst_ops->link_failure == NULL))
3915                         dst_ops->link_failure = xfrm_link_failure;
3916                 if (likely(dst_ops->neigh_lookup == NULL))
3917                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
3918                 if (likely(!dst_ops->confirm_neigh))
3919                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
3920                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
3921         }
3922         spin_unlock(&xfrm_policy_afinfo_lock);
3923
3924         return err;
3925 }
3926 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
3927
3928 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
3929 {
3930         struct dst_ops *dst_ops = afinfo->dst_ops;
3931         int i;
3932
3933         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
3934                 if (xfrm_policy_afinfo[i] != afinfo)
3935                         continue;
3936                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
3937                 break;
3938         }
3939
3940         synchronize_rcu();
3941
3942         dst_ops->kmem_cachep = NULL;
3943         dst_ops->check = NULL;
3944         dst_ops->negative_advice = NULL;
3945         dst_ops->link_failure = NULL;
3946 }
3947 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
3948
3949 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
3950 {
3951         spin_lock(&xfrm_if_cb_lock);
3952         rcu_assign_pointer(xfrm_if_cb, ifcb);
3953         spin_unlock(&xfrm_if_cb_lock);
3954 }
3955 EXPORT_SYMBOL(xfrm_if_register_cb);
3956
3957 void xfrm_if_unregister_cb(void)
3958 {
3959         RCU_INIT_POINTER(xfrm_if_cb, NULL);
3960         synchronize_rcu();
3961 }
3962 EXPORT_SYMBOL(xfrm_if_unregister_cb);
3963
3964 #ifdef CONFIG_XFRM_STATISTICS
3965 static int __net_init xfrm_statistics_init(struct net *net)
3966 {
3967         int rv;
3968         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
3969         if (!net->mib.xfrm_statistics)
3970                 return -ENOMEM;
3971         rv = xfrm_proc_init(net);
3972         if (rv < 0)
3973                 free_percpu(net->mib.xfrm_statistics);
3974         return rv;
3975 }
3976
3977 static void xfrm_statistics_fini(struct net *net)
3978 {
3979         xfrm_proc_fini(net);
3980         free_percpu(net->mib.xfrm_statistics);
3981 }
3982 #else
3983 static int __net_init xfrm_statistics_init(struct net *net)
3984 {
3985         return 0;
3986 }
3987
3988 static void xfrm_statistics_fini(struct net *net)
3989 {
3990 }
3991 #endif
3992
3993 static int __net_init xfrm_policy_init(struct net *net)
3994 {
3995         unsigned int hmask, sz;
3996         int dir, err;
3997
3998         if (net_eq(net, &init_net)) {
3999                 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
4000                                            sizeof(struct xfrm_dst),
4001                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4002                                            NULL);
4003                 err = rhashtable_init(&xfrm_policy_inexact_table,
4004                                       &xfrm_pol_inexact_params);
4005                 BUG_ON(err);
4006         }
4007
4008         hmask = 8 - 1;
4009         sz = (hmask+1) * sizeof(struct hlist_head);
4010
4011         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4012         if (!net->xfrm.policy_byidx)
4013                 goto out_byidx;
4014         net->xfrm.policy_idx_hmask = hmask;
4015
4016         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4017                 struct xfrm_policy_hash *htab;
4018
4019                 net->xfrm.policy_count[dir] = 0;
4020                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4021                 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]);
4022
4023                 htab = &net->xfrm.policy_bydst[dir];
4024                 htab->table = xfrm_hash_alloc(sz);
4025                 if (!htab->table)
4026                         goto out_bydst;
4027                 htab->hmask = hmask;
4028                 htab->dbits4 = 32;
4029                 htab->sbits4 = 32;
4030                 htab->dbits6 = 128;
4031                 htab->sbits6 = 128;
4032         }
4033         net->xfrm.policy_hthresh.lbits4 = 32;
4034         net->xfrm.policy_hthresh.rbits4 = 32;
4035         net->xfrm.policy_hthresh.lbits6 = 128;
4036         net->xfrm.policy_hthresh.rbits6 = 128;
4037
4038         seqlock_init(&net->xfrm.policy_hthresh.lock);
4039
4040         INIT_LIST_HEAD(&net->xfrm.policy_all);
4041         INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4042         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4043         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4044         return 0;
4045
4046 out_bydst:
4047         for (dir--; dir >= 0; dir--) {
4048                 struct xfrm_policy_hash *htab;
4049
4050                 htab = &net->xfrm.policy_bydst[dir];
4051                 xfrm_hash_free(htab->table, sz);
4052         }
4053         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4054 out_byidx:
4055         return -ENOMEM;
4056 }
4057
4058 static void xfrm_policy_fini(struct net *net)
4059 {
4060         struct xfrm_pol_inexact_bin *b, *t;
4061         unsigned int sz;
4062         int dir;
4063
4064         flush_work(&net->xfrm.policy_hash_work);
4065 #ifdef CONFIG_XFRM_SUB_POLICY
4066         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4067 #endif
4068         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4069
4070         WARN_ON(!list_empty(&net->xfrm.policy_all));
4071
4072         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4073                 struct xfrm_policy_hash *htab;
4074
4075                 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir]));
4076
4077                 htab = &net->xfrm.policy_bydst[dir];
4078                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4079                 WARN_ON(!hlist_empty(htab->table));
4080                 xfrm_hash_free(htab->table, sz);
4081         }
4082
4083         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4084         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4085         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4086
4087         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4088         list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4089                 __xfrm_policy_inexact_prune_bin(b, true);
4090         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4091 }
4092
4093 static int __net_init xfrm_net_init(struct net *net)
4094 {
4095         int rv;
4096
4097         /* Initialize the per-net locks here */
4098         spin_lock_init(&net->xfrm.xfrm_state_lock);
4099         spin_lock_init(&net->xfrm.xfrm_policy_lock);
4100         seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4101         mutex_init(&net->xfrm.xfrm_cfg_mutex);
4102
4103         rv = xfrm_statistics_init(net);
4104         if (rv < 0)
4105                 goto out_statistics;
4106         rv = xfrm_state_init(net);
4107         if (rv < 0)
4108                 goto out_state;
4109         rv = xfrm_policy_init(net);
4110         if (rv < 0)
4111                 goto out_policy;
4112         rv = xfrm_sysctl_init(net);
4113         if (rv < 0)
4114                 goto out_sysctl;
4115
4116         return 0;
4117
4118 out_sysctl:
4119         xfrm_policy_fini(net);
4120 out_policy:
4121         xfrm_state_fini(net);
4122 out_state:
4123         xfrm_statistics_fini(net);
4124 out_statistics:
4125         return rv;
4126 }
4127
4128 static void __net_exit xfrm_net_exit(struct net *net)
4129 {
4130         xfrm_sysctl_fini(net);
4131         xfrm_policy_fini(net);
4132         xfrm_state_fini(net);
4133         xfrm_statistics_fini(net);
4134 }
4135
4136 static struct pernet_operations __net_initdata xfrm_net_ops = {
4137         .init = xfrm_net_init,
4138         .exit = xfrm_net_exit,
4139 };
4140
4141 void __init xfrm_init(void)
4142 {
4143         register_pernet_subsys(&xfrm_net_ops);
4144         xfrm_dev_init();
4145         xfrm_input_init();
4146
4147 #ifdef CONFIG_XFRM_ESPINTCP
4148         espintcp_init();
4149 #endif
4150 }
4151
4152 #ifdef CONFIG_AUDITSYSCALL
4153 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4154                                          struct audit_buffer *audit_buf)
4155 {
4156         struct xfrm_sec_ctx *ctx = xp->security;
4157         struct xfrm_selector *sel = &xp->selector;
4158
4159         if (ctx)
4160                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4161                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4162
4163         switch (sel->family) {
4164         case AF_INET:
4165                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4166                 if (sel->prefixlen_s != 32)
4167                         audit_log_format(audit_buf, " src_prefixlen=%d",
4168                                          sel->prefixlen_s);
4169                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4170                 if (sel->prefixlen_d != 32)
4171                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4172                                          sel->prefixlen_d);
4173                 break;
4174         case AF_INET6:
4175                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4176                 if (sel->prefixlen_s != 128)
4177                         audit_log_format(audit_buf, " src_prefixlen=%d",
4178                                          sel->prefixlen_s);
4179                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4180                 if (sel->prefixlen_d != 128)
4181                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4182                                          sel->prefixlen_d);
4183                 break;
4184         }
4185 }
4186
4187 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4188 {
4189         struct audit_buffer *audit_buf;
4190
4191         audit_buf = xfrm_audit_start("SPD-add");
4192         if (audit_buf == NULL)
4193                 return;
4194         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4195         audit_log_format(audit_buf, " res=%u", result);
4196         xfrm_audit_common_policyinfo(xp, audit_buf);
4197         audit_log_end(audit_buf);
4198 }
4199 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4200
4201 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4202                               bool task_valid)
4203 {
4204         struct audit_buffer *audit_buf;
4205
4206         audit_buf = xfrm_audit_start("SPD-delete");
4207         if (audit_buf == NULL)
4208                 return;
4209         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4210         audit_log_format(audit_buf, " res=%u", result);
4211         xfrm_audit_common_policyinfo(xp, audit_buf);
4212         audit_log_end(audit_buf);
4213 }
4214 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4215 #endif
4216
4217 #ifdef CONFIG_XFRM_MIGRATE
4218 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp,
4219                                         const struct xfrm_selector *sel_tgt)
4220 {
4221         if (sel_cmp->proto == IPSEC_ULPROTO_ANY) {
4222                 if (sel_tgt->family == sel_cmp->family &&
4223                     xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr,
4224                                     sel_cmp->family) &&
4225                     xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr,
4226                                     sel_cmp->family) &&
4227                     sel_tgt->prefixlen_d == sel_cmp->prefixlen_d &&
4228                     sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) {
4229                         return true;
4230                 }
4231         } else {
4232                 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) {
4233                         return true;
4234                 }
4235         }
4236         return false;
4237 }
4238
4239 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4240                                                     u8 dir, u8 type, struct net *net)
4241 {
4242         struct xfrm_policy *pol, *ret = NULL;
4243         struct hlist_head *chain;
4244         u32 priority = ~0U;
4245
4246         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4247         chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir);
4248         hlist_for_each_entry(pol, chain, bydst) {
4249                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4250                     pol->type == type) {
4251                         ret = pol;
4252                         priority = ret->priority;
4253                         break;
4254                 }
4255         }
4256         chain = &net->xfrm.policy_inexact[dir];
4257         hlist_for_each_entry(pol, chain, bydst_inexact_list) {
4258                 if ((pol->priority >= priority) && ret)
4259                         break;
4260
4261                 if (xfrm_migrate_selector_match(sel, &pol->selector) &&
4262                     pol->type == type) {
4263                         ret = pol;
4264                         break;
4265                 }
4266         }
4267
4268         xfrm_pol_hold(ret);
4269
4270         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4271
4272         return ret;
4273 }
4274
4275 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4276 {
4277         int match = 0;
4278
4279         if (t->mode == m->mode && t->id.proto == m->proto &&
4280             (m->reqid == 0 || t->reqid == m->reqid)) {
4281                 switch (t->mode) {
4282                 case XFRM_MODE_TUNNEL:
4283                 case XFRM_MODE_BEET:
4284                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4285                                             m->old_family) &&
4286                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
4287                                             m->old_family)) {
4288                                 match = 1;
4289                         }
4290                         break;
4291                 case XFRM_MODE_TRANSPORT:
4292                         /* in case of transport mode, template does not store
4293                            any IP addresses, hence we just compare mode and
4294                            protocol */
4295                         match = 1;
4296                         break;
4297                 default:
4298                         break;
4299                 }
4300         }
4301         return match;
4302 }
4303
4304 /* update endpoint address(es) of template(s) */
4305 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4306                                struct xfrm_migrate *m, int num_migrate)
4307 {
4308         struct xfrm_migrate *mp;
4309         int i, j, n = 0;
4310
4311         write_lock_bh(&pol->lock);
4312         if (unlikely(pol->walk.dead)) {
4313                 /* target policy has been deleted */
4314                 write_unlock_bh(&pol->lock);
4315                 return -ENOENT;
4316         }
4317
4318         for (i = 0; i < pol->xfrm_nr; i++) {
4319                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4320                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4321                                 continue;
4322                         n++;
4323                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4324                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4325                                 continue;
4326                         /* update endpoints */
4327                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4328                                sizeof(pol->xfrm_vec[i].id.daddr));
4329                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4330                                sizeof(pol->xfrm_vec[i].saddr));
4331                         pol->xfrm_vec[i].encap_family = mp->new_family;
4332                         /* flush bundles */
4333                         atomic_inc(&pol->genid);
4334                 }
4335         }
4336
4337         write_unlock_bh(&pol->lock);
4338
4339         if (!n)
4340                 return -ENODATA;
4341
4342         return 0;
4343 }
4344
4345 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate)
4346 {
4347         int i, j;
4348
4349         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH)
4350                 return -EINVAL;
4351
4352         for (i = 0; i < num_migrate; i++) {
4353                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4354                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family))
4355                         return -EINVAL;
4356
4357                 /* check if there is any duplicated entry */
4358                 for (j = i + 1; j < num_migrate; j++) {
4359                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4360                                     sizeof(m[i].old_daddr)) &&
4361                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4362                                     sizeof(m[i].old_saddr)) &&
4363                             m[i].proto == m[j].proto &&
4364                             m[i].mode == m[j].mode &&
4365                             m[i].reqid == m[j].reqid &&
4366                             m[i].old_family == m[j].old_family)
4367                                 return -EINVAL;
4368                 }
4369         }
4370
4371         return 0;
4372 }
4373
4374 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4375                  struct xfrm_migrate *m, int num_migrate,
4376                  struct xfrm_kmaddress *k, struct net *net,
4377                  struct xfrm_encap_tmpl *encap)
4378 {
4379         int i, err, nx_cur = 0, nx_new = 0;
4380         struct xfrm_policy *pol = NULL;
4381         struct xfrm_state *x, *xc;
4382         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4383         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4384         struct xfrm_migrate *mp;
4385
4386         /* Stage 0 - sanity checks */
4387         if ((err = xfrm_migrate_check(m, num_migrate)) < 0)
4388                 goto out;
4389
4390         if (dir >= XFRM_POLICY_MAX) {
4391                 err = -EINVAL;
4392                 goto out;
4393         }
4394
4395         /* Stage 1 - find policy */
4396         if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) {
4397                 err = -ENOENT;
4398                 goto out;
4399         }
4400
4401         /* Stage 2 - find and update state(s) */
4402         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4403                 if ((x = xfrm_migrate_state_find(mp, net))) {
4404                         x_cur[nx_cur] = x;
4405                         nx_cur++;
4406                         xc = xfrm_state_migrate(x, mp, encap);
4407                         if (xc) {
4408                                 x_new[nx_new] = xc;
4409                                 nx_new++;
4410                         } else {
4411                                 err = -ENODATA;
4412                                 goto restore_state;
4413                         }
4414                 }
4415         }
4416
4417         /* Stage 3 - update policy */
4418         if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0)
4419                 goto restore_state;
4420
4421         /* Stage 4 - delete old state(s) */
4422         if (nx_cur) {
4423                 xfrm_states_put(x_cur, nx_cur);
4424                 xfrm_states_delete(x_cur, nx_cur);
4425         }
4426
4427         /* Stage 5 - announce */
4428         km_migrate(sel, dir, type, m, num_migrate, k, encap);
4429
4430         xfrm_pol_put(pol);
4431
4432         return 0;
4433 out:
4434         return err;
4435
4436 restore_state:
4437         if (pol)
4438                 xfrm_pol_put(pol);
4439         if (nx_cur)
4440                 xfrm_states_put(x_cur, nx_cur);
4441         if (nx_new)
4442                 xfrm_states_delete(x_new, nx_new);
4443
4444         return err;
4445 }
4446 EXPORT_SYMBOL(xfrm_migrate);
4447 #endif