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 <[email protected]>
   9  *              IPv6 support
  10  *      Kazunori MIYAZAWA @USAGI
  11  *      YOSHIFUJI Hideaki
  12  *              Split up af-specific portion
  13  *      Derek Atkins <[email protected]>          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 <linux/icmp.h>
  33 #include <net/dst.h>
  34 #include <net/flow.h>
  35 #include <net/inet_ecn.h>
  36 #include <net/xfrm.h>
  37 #include <net/ip.h>
  38 #include <net/gre.h>
  39 #if IS_ENABLED(CONFIG_IPV6_MIP6)
  40 #include <net/mip6.h>
  41 #endif
  42 #ifdef CONFIG_XFRM_STATISTICS
  43 #include <net/snmp.h>
  44 #endif
  45 #ifdef CONFIG_XFRM_ESPINTCP
  46 #include <net/espintcp.h>
  47 #endif
  48 #include <net/inet_dscp.h>
  49
  50 #include "xfrm_hash.h"
  51
  52 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10))
  53 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ))
  54 #define XFRM_MAX_QUEUE_LEN      100
  55
  56 struct xfrm_flo {
  57         struct dst_entry *dst_orig;
  58         u8 flags;
  59 };
  60
  61 /* prefixes smaller than this are stored in lists, not trees. */
  62 #define INEXACT_PREFIXLEN_IPV4  16
  63 #define INEXACT_PREFIXLEN_IPV6  48
  64
  65 struct xfrm_pol_inexact_node {
  66         struct rb_node node;
  67         union {
  68                 xfrm_address_t addr;
  69                 struct rcu_head rcu;
  70         };
  71         u8 prefixlen;
  72
  73         struct rb_root root;
  74
  75         /* the policies matching this node, can be empty list */
  76         struct hlist_head hhead;
  77 };
  78
  79 /* xfrm inexact policy search tree:
  80  * xfrm_pol_inexact_bin = hash(dir,type,family,if_id);
  81  *  |
  82  * +---- root_d: sorted by daddr:prefix
  83  * |                 |
  84  * |        xfrm_pol_inexact_node
  85  * |                 |
  86  * |                 +- root: sorted by saddr/prefix
  87  * |                 |              |
  88  * |                 |         xfrm_pol_inexact_node
  89  * |                 |              |
  90  * |                 |              + root: unused
  91  * |                 |              |
  92  * |                 |              + hhead: saddr:daddr policies
  93  * |                 |
  94  * |                 +- coarse policies and all any:daddr policies
  95  * |
  96  * +---- root_s: sorted by saddr:prefix
  97  * |                 |
  98  * |        xfrm_pol_inexact_node
  99  * |                 |
 100  * |                 + root: unused
 101  * |                 |
 102  * |                 + hhead: saddr:any policies
 103  * |
 104  * +---- coarse policies and all any:any policies
 105  *
 106  * Lookups return four candidate lists:
 107  * 1. any:any list from top-level xfrm_pol_inexact_bin
 108  * 2. any:daddr list from daddr tree
 109  * 3. saddr:daddr list from 2nd level daddr tree
 110  * 4. saddr:any list from saddr tree
 111  *
 112  * This result set then needs to be searched for the policy with
 113  * the lowest priority.  If two candidates have the same priority, the
 114  * struct xfrm_policy pos member with the lower number is used.
 115  *
 116  * This replicates previous single-list-search algorithm which would
 117  * return first matching policy in the (ordered-by-priority) list.
 118  */
 119
 120 struct xfrm_pol_inexact_key {
 121         possible_net_t net;
 122         u32 if_id;
 123         u16 family;
 124         u8 dir, type;
 125 };
 126
 127 struct xfrm_pol_inexact_bin {
 128         struct xfrm_pol_inexact_key k;
 129         struct rhash_head head;
 130         /* list containing '*:*' policies */
 131         struct hlist_head hhead;
 132
 133         seqcount_spinlock_t count;
 134         /* tree sorted by daddr/prefix */
 135         struct rb_root root_d;
 136
 137         /* tree sorted by saddr/prefix */
 138         struct rb_root root_s;
 139
 140         /* slow path below */
 141         struct list_head inexact_bins;
 142         struct rcu_head rcu;
 143 };
 144
 145 enum xfrm_pol_inexact_candidate_type {
 146         XFRM_POL_CAND_BOTH,
 147         XFRM_POL_CAND_SADDR,
 148         XFRM_POL_CAND_DADDR,
 149         XFRM_POL_CAND_ANY,
 150
 151         XFRM_POL_CAND_MAX,
 152 };
 153
 154 struct xfrm_pol_inexact_candidates {
 155         struct hlist_head *res[XFRM_POL_CAND_MAX];
 156 };
 157
 158 struct xfrm_flow_keys {
 159         struct flow_dissector_key_basic basic;
 160         struct flow_dissector_key_control control;
 161         union {
 162                 struct flow_dissector_key_ipv4_addrs ipv4;
 163                 struct flow_dissector_key_ipv6_addrs ipv6;
 164         } addrs;
 165         struct flow_dissector_key_ip ip;
 166         struct flow_dissector_key_icmp icmp;
 167         struct flow_dissector_key_ports ports;
 168         struct flow_dissector_key_keyid gre;
 169 };
 170
 171 static struct flow_dissector xfrm_session_dissector __ro_after_init;
 172
 173 static DEFINE_SPINLOCK(xfrm_if_cb_lock);
 174 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly;
 175
 176 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock);
 177 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1]
 178                                                 __read_mostly;
 179
 180 static struct kmem_cache *xfrm_dst_cache __ro_after_init;
 181
 182 static struct rhashtable xfrm_policy_inexact_table;
 183 static const struct rhashtable_params xfrm_pol_inexact_params;
 184
 185 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr);
 186 static int stale_bundle(struct dst_entry *dst);
 187 static int xfrm_bundle_ok(struct xfrm_dst *xdst);
 188 static void xfrm_policy_queue_process(struct timer_list *t);
 189
 190 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir);
 191 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
 192                                                 int dir);
 193
 194 static struct xfrm_pol_inexact_bin *
 195 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir,
 196                            u32 if_id);
 197
 198 static struct xfrm_pol_inexact_bin *
 199 xfrm_policy_inexact_lookup_rcu(struct net *net,
 200                                u8 type, u16 family, u8 dir, u32 if_id);
 201 static struct xfrm_policy *
 202 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy,
 203                         bool excl);
 204
 205 static bool
 206 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
 207                                     struct xfrm_pol_inexact_bin *b,
 208                                     const xfrm_address_t *saddr,
 209                                     const xfrm_address_t *daddr);
 210
 211 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy)
 212 {
 213         return refcount_inc_not_zero(&policy->refcnt);
 214 }
 215
 216 static inline bool
 217 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 218 {
 219         const struct flowi4 *fl4 = &fl->u.ip4;
 220
 221         return  addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) &&
 222                 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) &&
 223                 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) &&
 224                 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) &&
 225                 (fl4->flowi4_proto == sel->proto || !sel->proto) &&
 226                 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex);
 227 }
 228
 229 static inline bool
 230 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl)
 231 {
 232         const struct flowi6 *fl6 = &fl->u.ip6;
 233
 234         return  addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) &&
 235                 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) &&
 236                 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) &&
 237                 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) &&
 238                 (fl6->flowi6_proto == sel->proto || !sel->proto) &&
 239                 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex);
 240 }
 241
 242 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl,
 243                          unsigned short family)
 244 {
 245         switch (family) {
 246         case AF_INET:
 247                 return __xfrm4_selector_match(sel, fl);
 248         case AF_INET6:
 249                 return __xfrm6_selector_match(sel, fl);
 250         }
 251         return false;
 252 }
 253
 254 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
 255 {
 256         const struct xfrm_policy_afinfo *afinfo;
 257
 258         if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
 259                 return NULL;
 260         rcu_read_lock();
 261         afinfo = rcu_dereference(xfrm_policy_afinfo[family]);
 262         if (unlikely(!afinfo))
 263                 rcu_read_unlock();
 264         return afinfo;
 265 }
 266
 267 /* Called with rcu_read_lock(). */
 268 static const struct xfrm_if_cb *xfrm_if_get_cb(void)
 269 {
 270         return rcu_dereference(xfrm_if_cb);
 271 }
 272
 273 struct dst_entry *__xfrm_dst_lookup(int family,
 274                                     const struct xfrm_dst_lookup_params *params)
 275 {
 276         const struct xfrm_policy_afinfo *afinfo;
 277         struct dst_entry *dst;
 278
 279         afinfo = xfrm_policy_get_afinfo(family);
 280         if (unlikely(afinfo == NULL))
 281                 return ERR_PTR(-EAFNOSUPPORT);
 282
 283         dst = afinfo->dst_lookup(params);
 284
 285         rcu_read_unlock();
 286
 287         return dst;
 288 }
 289 EXPORT_SYMBOL(__xfrm_dst_lookup);
 290
 291 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x,
 292                                                 dscp_t dscp, int oif,
 293                                                 xfrm_address_t *prev_saddr,
 294                                                 xfrm_address_t *prev_daddr,
 295                                                 int family, u32 mark)
 296 {
 297         struct xfrm_dst_lookup_params params;
 298         struct net *net = xs_net(x);
 299         xfrm_address_t *saddr = &x->props.saddr;
 300         xfrm_address_t *daddr = &x->id.daddr;
 301         struct dst_entry *dst;
 302
 303         if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) {
 304                 saddr = x->coaddr;
 305                 daddr = prev_daddr;
 306         }
 307         if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) {
 308                 saddr = prev_saddr;
 309                 daddr = x->coaddr;
 310         }
 311
 312         params.net = net;
 313         params.saddr = saddr;
 314         params.daddr = daddr;
 315         params.dscp = dscp;
 316         params.oif = oif;
 317         params.mark = mark;
 318         params.ipproto = x->id.proto;
 319         if (x->encap) {
 320                 switch (x->encap->encap_type) {
 321                 case UDP_ENCAP_ESPINUDP:
 322                         params.ipproto = IPPROTO_UDP;
 323                         params.uli.ports.sport = x->encap->encap_sport;
 324                         params.uli.ports.dport = x->encap->encap_dport;
 325                         break;
 326                 case TCP_ENCAP_ESPINTCP:
 327                         params.ipproto = IPPROTO_TCP;
 328                         params.uli.ports.sport = x->encap->encap_sport;
 329                         params.uli.ports.dport = x->encap->encap_dport;
 330                         break;
 331                 }
 332         }
 333
 334         dst = __xfrm_dst_lookup(family, &params);
 335
 336         if (!IS_ERR(dst)) {
 337                 if (prev_saddr != saddr)
 338                         memcpy(prev_saddr, saddr,  sizeof(*prev_saddr));
 339                 if (prev_daddr != daddr)
 340                         memcpy(prev_daddr, daddr,  sizeof(*prev_daddr));
 341         }
 342
 343         return dst;
 344 }
 345
 346 static inline unsigned long make_jiffies(long secs)
 347 {
 348         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
 349                 return MAX_SCHEDULE_TIMEOUT-1;
 350         else
 351                 return secs*HZ;
 352 }
 353
 354 static void xfrm_policy_timer(struct timer_list *t)
 355 {
 356         struct xfrm_policy *xp = from_timer(xp, t, timer);
 357         time64_t now = ktime_get_real_seconds();
 358         time64_t next = TIME64_MAX;
 359         int warn = 0;
 360         int dir;
 361
 362         read_lock(&xp->lock);
 363
 364         if (unlikely(xp->walk.dead))
 365                 goto out;
 366
 367         dir = xfrm_policy_id2dir(xp->index);
 368
 369         if (xp->lft.hard_add_expires_seconds) {
 370                 time64_t tmo = xp->lft.hard_add_expires_seconds +
 371                         xp->curlft.add_time - now;
 372                 if (tmo <= 0)
 373                         goto expired;
 374                 if (tmo < next)
 375                         next = tmo;
 376         }
 377         if (xp->lft.hard_use_expires_seconds) {
 378                 time64_t tmo = xp->lft.hard_use_expires_seconds +
 379                         (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
 380                 if (tmo <= 0)
 381                         goto expired;
 382                 if (tmo < next)
 383                         next = tmo;
 384         }
 385         if (xp->lft.soft_add_expires_seconds) {
 386                 time64_t tmo = xp->lft.soft_add_expires_seconds +
 387                         xp->curlft.add_time - now;
 388                 if (tmo <= 0) {
 389                         warn = 1;
 390                         tmo = XFRM_KM_TIMEOUT;
 391                 }
 392                 if (tmo < next)
 393                         next = tmo;
 394         }
 395         if (xp->lft.soft_use_expires_seconds) {
 396                 time64_t tmo = xp->lft.soft_use_expires_seconds +
 397                         (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now;
 398                 if (tmo <= 0) {
 399                         warn = 1;
 400                         tmo = XFRM_KM_TIMEOUT;
 401                 }
 402                 if (tmo < next)
 403                         next = tmo;
 404         }
 405
 406         if (warn)
 407                 km_policy_expired(xp, dir, 0, 0);
 408         if (next != TIME64_MAX &&
 409             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
 410                 xfrm_pol_hold(xp);
 411
 412 out:
 413         read_unlock(&xp->lock);
 414         xfrm_pol_put(xp);
 415         return;
 416
 417 expired:
 418         read_unlock(&xp->lock);
 419         if (!xfrm_policy_delete(xp, dir))
 420                 km_policy_expired(xp, dir, 1, 0);
 421         xfrm_pol_put(xp);
 422 }
 423
 424 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
 425  * SPD calls.
 426  */
 427
 428 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp)
 429 {
 430         struct xfrm_policy *policy;
 431
 432         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 433
 434         if (policy) {
 435                 write_pnet(&policy->xp_net, net);
 436                 INIT_LIST_HEAD(&policy->walk.all);
 437                 INIT_HLIST_HEAD(&policy->state_cache_list);
 438                 INIT_HLIST_NODE(&policy->bydst);
 439                 INIT_HLIST_NODE(&policy->byidx);
 440                 rwlock_init(&policy->lock);
 441                 refcount_set(&policy->refcnt, 1);
 442                 skb_queue_head_init(&policy->polq.hold_queue);
 443                 timer_setup(&policy->timer, xfrm_policy_timer, 0);
 444                 timer_setup(&policy->polq.hold_timer,
 445                             xfrm_policy_queue_process, 0);
 446         }
 447         return policy;
 448 }
 449 EXPORT_SYMBOL(xfrm_policy_alloc);
 450
 451 static void xfrm_policy_destroy_rcu(struct rcu_head *head)
 452 {
 453         struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu);
 454
 455         security_xfrm_policy_free(policy->security);
 456         kfree(policy);
 457 }
 458
 459 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
 460
 461 void xfrm_policy_destroy(struct xfrm_policy *policy)
 462 {
 463         BUG_ON(!policy->walk.dead);
 464
 465         if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer))
 466                 BUG();
 467
 468         xfrm_dev_policy_free(policy);
 469         call_rcu(&policy->rcu, xfrm_policy_destroy_rcu);
 470 }
 471 EXPORT_SYMBOL(xfrm_policy_destroy);
 472
 473 /* Rule must be locked. Release descendant resources, announce
 474  * entry dead. The rule must be unlinked from lists to the moment.
 475  */
 476
 477 static void xfrm_policy_kill(struct xfrm_policy *policy)
 478 {
 479         struct net *net = xp_net(policy);
 480         struct xfrm_state *x;
 481
 482         xfrm_dev_policy_delete(policy);
 483
 484         write_lock_bh(&policy->lock);
 485         policy->walk.dead = 1;
 486         write_unlock_bh(&policy->lock);
 487
 488         atomic_inc(&policy->genid);
 489
 490         if (del_timer(&policy->polq.hold_timer))
 491                 xfrm_pol_put(policy);
 492         skb_queue_purge(&policy->polq.hold_queue);
 493
 494         if (del_timer(&policy->timer))
 495                 xfrm_pol_put(policy);
 496
 497         /* XXX: Flush state cache */
 498         spin_lock_bh(&net->xfrm.xfrm_state_lock);
 499         hlist_for_each_entry_rcu(x, &policy->state_cache_list, state_cache) {
 500                 hlist_del_init_rcu(&x->state_cache);
 501         }
 502         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 503
 504         xfrm_pol_put(policy);
 505 }
 506
 507 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
 508
 509 static inline unsigned int idx_hash(struct net *net, u32 index)
 510 {
 511         return __idx_hash(index, net->xfrm.policy_idx_hmask);
 512 }
 513
 514 /* calculate policy hash thresholds */
 515 static void __get_hash_thresh(struct net *net,
 516                               unsigned short family, int dir,
 517                               u8 *dbits, u8 *sbits)
 518 {
 519         switch (family) {
 520         case AF_INET:
 521                 *dbits = net->xfrm.policy_bydst[dir].dbits4;
 522                 *sbits = net->xfrm.policy_bydst[dir].sbits4;
 523                 break;
 524
 525         case AF_INET6:
 526                 *dbits = net->xfrm.policy_bydst[dir].dbits6;
 527                 *sbits = net->xfrm.policy_bydst[dir].sbits6;
 528                 break;
 529
 530         default:
 531                 *dbits = 0;
 532                 *sbits = 0;
 533         }
 534 }
 535
 536 static struct hlist_head *policy_hash_bysel(struct net *net,
 537                                             const struct xfrm_selector *sel,
 538                                             unsigned short family, int dir)
 539 {
 540         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 541         unsigned int hash;
 542         u8 dbits;
 543         u8 sbits;
 544
 545         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 546         hash = __sel_hash(sel, family, hmask, dbits, sbits);
 547
 548         if (hash == hmask + 1)
 549                 return NULL;
 550
 551         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 552                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 553 }
 554
 555 static struct hlist_head *policy_hash_direct(struct net *net,
 556                                              const xfrm_address_t *daddr,
 557                                              const xfrm_address_t *saddr,
 558                                              unsigned short family, int dir)
 559 {
 560         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 561         unsigned int hash;
 562         u8 dbits;
 563         u8 sbits;
 564
 565         __get_hash_thresh(net, family, dir, &dbits, &sbits);
 566         hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits);
 567
 568         return rcu_dereference_check(net->xfrm.policy_bydst[dir].table,
 569                      lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash;
 570 }
 571
 572 static void xfrm_dst_hash_transfer(struct net *net,
 573                                    struct hlist_head *list,
 574                                    struct hlist_head *ndsttable,
 575                                    unsigned int nhashmask,
 576                                    int dir)
 577 {
 578         struct hlist_node *tmp, *entry0 = NULL;
 579         struct xfrm_policy *pol;
 580         unsigned int h0 = 0;
 581         u8 dbits;
 582         u8 sbits;
 583
 584 redo:
 585         hlist_for_each_entry_safe(pol, tmp, list, bydst) {
 586                 unsigned int h;
 587
 588                 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits);
 589                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
 590                                 pol->family, nhashmask, dbits, sbits);
 591                 if (!entry0 || pol->xdo.type == XFRM_DEV_OFFLOAD_PACKET) {
 592                         hlist_del_rcu(&pol->bydst);
 593                         hlist_add_head_rcu(&pol->bydst, ndsttable + h);
 594                         h0 = h;
 595                 } else {
 596                         if (h != h0)
 597                                 continue;
 598                         hlist_del_rcu(&pol->bydst);
 599                         hlist_add_behind_rcu(&pol->bydst, entry0);
 600                 }
 601                 entry0 = &pol->bydst;
 602         }
 603         if (!hlist_empty(list)) {
 604                 entry0 = NULL;
 605                 goto redo;
 606         }
 607 }
 608
 609 static void xfrm_idx_hash_transfer(struct hlist_head *list,
 610                                    struct hlist_head *nidxtable,
 611                                    unsigned int nhashmask)
 612 {
 613         struct hlist_node *tmp;
 614         struct xfrm_policy *pol;
 615
 616         hlist_for_each_entry_safe(pol, tmp, list, byidx) {
 617                 unsigned int h;
 618
 619                 h = __idx_hash(pol->index, nhashmask);
 620                 hlist_add_head(&pol->byidx, nidxtable+h);
 621         }
 622 }
 623
 624 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
 625 {
 626         return ((old_hmask + 1) << 1) - 1;
 627 }
 628
 629 static void xfrm_bydst_resize(struct net *net, int dir)
 630 {
 631         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 632         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 633         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 634         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
 635         struct hlist_head *odst;
 636         int i;
 637
 638         if (!ndst)
 639                 return;
 640
 641         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 642         write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
 643
 644         odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table,
 645                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
 646
 647         for (i = hmask; i >= 0; i--)
 648                 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir);
 649
 650         rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst);
 651         net->xfrm.policy_bydst[dir].hmask = nhashmask;
 652
 653         write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
 654         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 655
 656         synchronize_rcu();
 657
 658         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
 659 }
 660
 661 static void xfrm_byidx_resize(struct net *net)
 662 {
 663         unsigned int hmask = net->xfrm.policy_idx_hmask;
 664         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
 665         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
 666         struct hlist_head *oidx = net->xfrm.policy_byidx;
 667         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
 668         int i;
 669
 670         if (!nidx)
 671                 return;
 672
 673         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
 674
 675         for (i = hmask; i >= 0; i--)
 676                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
 677
 678         net->xfrm.policy_byidx = nidx;
 679         net->xfrm.policy_idx_hmask = nhashmask;
 680
 681         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
 682
 683         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
 684 }
 685
 686 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total)
 687 {
 688         unsigned int cnt = net->xfrm.policy_count[dir];
 689         unsigned int hmask = net->xfrm.policy_bydst[dir].hmask;
 690
 691         if (total)
 692                 *total += cnt;
 693
 694         if ((hmask + 1) < xfrm_policy_hashmax &&
 695             cnt > hmask)
 696                 return 1;
 697
 698         return 0;
 699 }
 700
 701 static inline int xfrm_byidx_should_resize(struct net *net, int total)
 702 {
 703         unsigned int hmask = net->xfrm.policy_idx_hmask;
 704
 705         if ((hmask + 1) < xfrm_policy_hashmax &&
 706             total > hmask)
 707                 return 1;
 708
 709         return 0;
 710 }
 711
 712 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si)
 713 {
 714         si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN];
 715         si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT];
 716         si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD];
 717         si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX];
 718         si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX];
 719         si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX];
 720         si->spdhcnt = net->xfrm.policy_idx_hmask;
 721         si->spdhmcnt = xfrm_policy_hashmax;
 722 }
 723 EXPORT_SYMBOL(xfrm_spd_getinfo);
 724
 725 static DEFINE_MUTEX(hash_resize_mutex);
 726 static void xfrm_hash_resize(struct work_struct *work)
 727 {
 728         struct net *net = container_of(work, struct net, xfrm.policy_hash_work);
 729         int dir, total;
 730
 731         mutex_lock(&hash_resize_mutex);
 732
 733         total = 0;
 734         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
 735                 if (xfrm_bydst_should_resize(net, dir, &total))
 736                         xfrm_bydst_resize(net, dir);
 737         }
 738         if (xfrm_byidx_should_resize(net, total))
 739                 xfrm_byidx_resize(net);
 740
 741         mutex_unlock(&hash_resize_mutex);
 742 }
 743
 744 /* Make sure *pol can be inserted into fastbin.
 745  * Useful to check that later insert requests will be successful
 746  * (provided xfrm_policy_lock is held throughout).
 747  */
 748 static struct xfrm_pol_inexact_bin *
 749 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir)
 750 {
 751         struct xfrm_pol_inexact_bin *bin, *prev;
 752         struct xfrm_pol_inexact_key k = {
 753                 .family = pol->family,
 754                 .type = pol->type,
 755                 .dir = dir,
 756                 .if_id = pol->if_id,
 757         };
 758         struct net *net = xp_net(pol);
 759
 760         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
 761
 762         write_pnet(&k.net, net);
 763         bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k,
 764                                      xfrm_pol_inexact_params);
 765         if (bin)
 766                 return bin;
 767
 768         bin = kzalloc(sizeof(*bin), GFP_ATOMIC);
 769         if (!bin)
 770                 return NULL;
 771
 772         bin->k = k;
 773         INIT_HLIST_HEAD(&bin->hhead);
 774         bin->root_d = RB_ROOT;
 775         bin->root_s = RB_ROOT;
 776         seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock);
 777
 778         prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table,
 779                                                 &bin->k, &bin->head,
 780                                                 xfrm_pol_inexact_params);
 781         if (!prev) {
 782                 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins);
 783                 return bin;
 784         }
 785
 786         kfree(bin);
 787
 788         return IS_ERR(prev) ? NULL : prev;
 789 }
 790
 791 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr,
 792                                                int family, u8 prefixlen)
 793 {
 794         if (xfrm_addr_any(addr, family))
 795                 return true;
 796
 797         if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6)
 798                 return true;
 799
 800         if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4)
 801                 return true;
 802
 803         return false;
 804 }
 805
 806 static bool
 807 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy)
 808 {
 809         const xfrm_address_t *addr;
 810         bool saddr_any, daddr_any;
 811         u8 prefixlen;
 812
 813         addr = &policy->selector.saddr;
 814         prefixlen = policy->selector.prefixlen_s;
 815
 816         saddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 817                                                        policy->family,
 818                                                        prefixlen);
 819         addr = &policy->selector.daddr;
 820         prefixlen = policy->selector.prefixlen_d;
 821         daddr_any = xfrm_pol_inexact_addr_use_any_list(addr,
 822                                                        policy->family,
 823                                                        prefixlen);
 824         return saddr_any && daddr_any;
 825 }
 826
 827 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node,
 828                                        const xfrm_address_t *addr, u8 prefixlen)
 829 {
 830         node->addr = *addr;
 831         node->prefixlen = prefixlen;
 832 }
 833
 834 static struct xfrm_pol_inexact_node *
 835 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen)
 836 {
 837         struct xfrm_pol_inexact_node *node;
 838
 839         node = kzalloc(sizeof(*node), GFP_ATOMIC);
 840         if (node)
 841                 xfrm_pol_inexact_node_init(node, addr, prefixlen);
 842
 843         return node;
 844 }
 845
 846 static int xfrm_policy_addr_delta(const xfrm_address_t *a,
 847                                   const xfrm_address_t *b,
 848                                   u8 prefixlen, u16 family)
 849 {
 850         u32 ma, mb, mask;
 851         unsigned int pdw, pbi;
 852         int delta = 0;
 853
 854         switch (family) {
 855         case AF_INET:
 856                 if (prefixlen == 0)
 857                         return 0;
 858                 mask = ~0U << (32 - prefixlen);
 859                 ma = ntohl(a->a4) & mask;
 860                 mb = ntohl(b->a4) & mask;
 861                 if (ma < mb)
 862                         delta = -1;
 863                 else if (ma > mb)
 864                         delta = 1;
 865                 break;
 866         case AF_INET6:
 867                 pdw = prefixlen >> 5;
 868                 pbi = prefixlen & 0x1f;
 869
 870                 if (pdw) {
 871                         delta = memcmp(a->a6, b->a6, pdw << 2);
 872                         if (delta)
 873                                 return delta;
 874                 }
 875                 if (pbi) {
 876                         mask = ~0U << (32 - pbi);
 877                         ma = ntohl(a->a6[pdw]) & mask;
 878                         mb = ntohl(b->a6[pdw]) & mask;
 879                         if (ma < mb)
 880                                 delta = -1;
 881                         else if (ma > mb)
 882                                 delta = 1;
 883                 }
 884                 break;
 885         default:
 886                 break;
 887         }
 888
 889         return delta;
 890 }
 891
 892 static void xfrm_policy_inexact_list_reinsert(struct net *net,
 893                                               struct xfrm_pol_inexact_node *n,
 894                                               u16 family)
 895 {
 896         unsigned int matched_s, matched_d;
 897         struct xfrm_policy *policy, *p;
 898
 899         matched_s = 0;
 900         matched_d = 0;
 901
 902         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
 903                 struct hlist_node *newpos = NULL;
 904                 bool matches_s, matches_d;
 905
 906                 if (policy->walk.dead || !policy->bydst_reinsert)
 907                         continue;
 908
 909                 WARN_ON_ONCE(policy->family != family);
 910
 911                 policy->bydst_reinsert = false;
 912                 hlist_for_each_entry(p, &n->hhead, bydst) {
 913                         if (policy->priority > p->priority)
 914                                 newpos = &p->bydst;
 915                         else if (policy->priority == p->priority &&
 916                                  policy->pos > p->pos)
 917                                 newpos = &p->bydst;
 918                         else
 919                                 break;
 920                 }
 921
 922                 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
 923                         hlist_add_behind_rcu(&policy->bydst, newpos);
 924                 else
 925                         hlist_add_head_rcu(&policy->bydst, &n->hhead);
 926
 927                 /* paranoia checks follow.
 928                  * Check that the reinserted policy matches at least
 929                  * saddr or daddr for current node prefix.
 930                  *
 931                  * Matching both is fine, matching saddr in one policy
 932                  * (but not daddr) and then matching only daddr in another
 933                  * is a bug.
 934                  */
 935                 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr,
 936                                                    &n->addr,
 937                                                    n->prefixlen,
 938                                                    family) == 0;
 939                 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr,
 940                                                    &n->addr,
 941                                                    n->prefixlen,
 942                                                    family) == 0;
 943                 if (matches_s && matches_d)
 944                         continue;
 945
 946                 WARN_ON_ONCE(!matches_s && !matches_d);
 947                 if (matches_s)
 948                         matched_s++;
 949                 if (matches_d)
 950                         matched_d++;
 951                 WARN_ON_ONCE(matched_s && matched_d);
 952         }
 953 }
 954
 955 static void xfrm_policy_inexact_node_reinsert(struct net *net,
 956                                               struct xfrm_pol_inexact_node *n,
 957                                               struct rb_root *new,
 958                                               u16 family)
 959 {
 960         struct xfrm_pol_inexact_node *node;
 961         struct rb_node **p, *parent;
 962
 963         /* we should not have another subtree here */
 964         WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root));
 965 restart:
 966         parent = NULL;
 967         p = &new->rb_node;
 968         while (*p) {
 969                 u8 prefixlen;
 970                 int delta;
 971
 972                 parent = *p;
 973                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
 974
 975                 prefixlen = min(node->prefixlen, n->prefixlen);
 976
 977                 delta = xfrm_policy_addr_delta(&n->addr, &node->addr,
 978                                                prefixlen, family);
 979                 if (delta < 0) {
 980                         p = &parent->rb_left;
 981                 } else if (delta > 0) {
 982                         p = &parent->rb_right;
 983                 } else {
 984                         bool same_prefixlen = node->prefixlen == n->prefixlen;
 985                         struct xfrm_policy *tmp;
 986
 987                         hlist_for_each_entry(tmp, &n->hhead, bydst) {
 988                                 tmp->bydst_reinsert = true;
 989                                 hlist_del_rcu(&tmp->bydst);
 990                         }
 991
 992                         node->prefixlen = prefixlen;
 993
 994                         xfrm_policy_inexact_list_reinsert(net, node, family);
 995
 996                         if (same_prefixlen) {
 997                                 kfree_rcu(n, rcu);
 998                                 return;
 999                         }
1000
1001                         rb_erase(*p, new);
1002                         kfree_rcu(n, rcu);
1003                         n = node;
1004                         goto restart;
1005                 }
1006         }
1007
1008         rb_link_node_rcu(&n->node, parent, p);
1009         rb_insert_color(&n->node, new);
1010 }
1011
1012 /* merge nodes v and n */
1013 static void xfrm_policy_inexact_node_merge(struct net *net,
1014                                            struct xfrm_pol_inexact_node *v,
1015                                            struct xfrm_pol_inexact_node *n,
1016                                            u16 family)
1017 {
1018         struct xfrm_pol_inexact_node *node;
1019         struct xfrm_policy *tmp;
1020         struct rb_node *rnode;
1021
1022         /* To-be-merged node v has a subtree.
1023          *
1024          * Dismantle it and insert its nodes to n->root.
1025          */
1026         while ((rnode = rb_first(&v->root)) != NULL) {
1027                 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node);
1028                 rb_erase(&node->node, &v->root);
1029                 xfrm_policy_inexact_node_reinsert(net, node, &n->root,
1030                                                   family);
1031         }
1032
1033         hlist_for_each_entry(tmp, &v->hhead, bydst) {
1034                 tmp->bydst_reinsert = true;
1035                 hlist_del_rcu(&tmp->bydst);
1036         }
1037
1038         xfrm_policy_inexact_list_reinsert(net, n, family);
1039 }
1040
1041 static struct xfrm_pol_inexact_node *
1042 xfrm_policy_inexact_insert_node(struct net *net,
1043                                 struct rb_root *root,
1044                                 xfrm_address_t *addr,
1045                                 u16 family, u8 prefixlen, u8 dir)
1046 {
1047         struct xfrm_pol_inexact_node *cached = NULL;
1048         struct rb_node **p, *parent = NULL;
1049         struct xfrm_pol_inexact_node *node;
1050
1051         p = &root->rb_node;
1052         while (*p) {
1053                 int delta;
1054
1055                 parent = *p;
1056                 node = rb_entry(*p, struct xfrm_pol_inexact_node, node);
1057
1058                 delta = xfrm_policy_addr_delta(addr, &node->addr,
1059                                                node->prefixlen,
1060                                                family);
1061                 if (delta == 0 && prefixlen >= node->prefixlen) {
1062                         WARN_ON_ONCE(cached); /* ipsec policies got lost */
1063                         return node;
1064                 }
1065
1066                 if (delta < 0)
1067                         p = &parent->rb_left;
1068                 else
1069                         p = &parent->rb_right;
1070
1071                 if (prefixlen < node->prefixlen) {
1072                         delta = xfrm_policy_addr_delta(addr, &node->addr,
1073                                                        prefixlen,
1074                                                        family);
1075                         if (delta)
1076                                 continue;
1077
1078                         /* This node is a subnet of the new prefix. It needs
1079                          * to be removed and re-inserted with the smaller
1080                          * prefix and all nodes that are now also covered
1081                          * by the reduced prefixlen.
1082                          */
1083                         rb_erase(&node->node, root);
1084
1085                         if (!cached) {
1086                                 xfrm_pol_inexact_node_init(node, addr,
1087                                                            prefixlen);
1088                                 cached = node;
1089                         } else {
1090                                 /* This node also falls within the new
1091                                  * prefixlen. Merge the to-be-reinserted
1092                                  * node and this one.
1093                                  */
1094                                 xfrm_policy_inexact_node_merge(net, node,
1095                                                                cached, family);
1096                                 kfree_rcu(node, rcu);
1097                         }
1098
1099                         /* restart */
1100                         p = &root->rb_node;
1101                         parent = NULL;
1102                 }
1103         }
1104
1105         node = cached;
1106         if (!node) {
1107                 node = xfrm_pol_inexact_node_alloc(addr, prefixlen);
1108                 if (!node)
1109                         return NULL;
1110         }
1111
1112         rb_link_node_rcu(&node->node, parent, p);
1113         rb_insert_color(&node->node, root);
1114
1115         return node;
1116 }
1117
1118 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm)
1119 {
1120         struct xfrm_pol_inexact_node *node;
1121         struct rb_node *rn = rb_first(r);
1122
1123         while (rn) {
1124                 node = rb_entry(rn, struct xfrm_pol_inexact_node, node);
1125
1126                 xfrm_policy_inexact_gc_tree(&node->root, rm);
1127                 rn = rb_next(rn);
1128
1129                 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) {
1130                         WARN_ON_ONCE(rm);
1131                         continue;
1132                 }
1133
1134                 rb_erase(&node->node, r);
1135                 kfree_rcu(node, rcu);
1136         }
1137 }
1138
1139 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit)
1140 {
1141         write_seqcount_begin(&b->count);
1142         xfrm_policy_inexact_gc_tree(&b->root_d, net_exit);
1143         xfrm_policy_inexact_gc_tree(&b->root_s, net_exit);
1144         write_seqcount_end(&b->count);
1145
1146         if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) ||
1147             !hlist_empty(&b->hhead)) {
1148                 WARN_ON_ONCE(net_exit);
1149                 return;
1150         }
1151
1152         if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head,
1153                                    xfrm_pol_inexact_params) == 0) {
1154                 list_del(&b->inexact_bins);
1155                 kfree_rcu(b, rcu);
1156         }
1157 }
1158
1159 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b)
1160 {
1161         struct net *net = read_pnet(&b->k.net);
1162
1163         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1164         __xfrm_policy_inexact_prune_bin(b, false);
1165         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1166 }
1167
1168 static void __xfrm_policy_inexact_flush(struct net *net)
1169 {
1170         struct xfrm_pol_inexact_bin *bin, *t;
1171
1172         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1173
1174         list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins)
1175                 __xfrm_policy_inexact_prune_bin(bin, false);
1176 }
1177
1178 static struct hlist_head *
1179 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin,
1180                                 struct xfrm_policy *policy, u8 dir)
1181 {
1182         struct xfrm_pol_inexact_node *n;
1183         struct net *net;
1184
1185         net = xp_net(policy);
1186         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1187
1188         if (xfrm_policy_inexact_insert_use_any_list(policy))
1189                 return &bin->hhead;
1190
1191         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr,
1192                                                policy->family,
1193                                                policy->selector.prefixlen_d)) {
1194                 write_seqcount_begin(&bin->count);
1195                 n = xfrm_policy_inexact_insert_node(net,
1196                                                     &bin->root_s,
1197                                                     &policy->selector.saddr,
1198                                                     policy->family,
1199                                                     policy->selector.prefixlen_s,
1200                                                     dir);
1201                 write_seqcount_end(&bin->count);
1202                 if (!n)
1203                         return NULL;
1204
1205                 return &n->hhead;
1206         }
1207
1208         /* daddr is fixed */
1209         write_seqcount_begin(&bin->count);
1210         n = xfrm_policy_inexact_insert_node(net,
1211                                             &bin->root_d,
1212                                             &policy->selector.daddr,
1213                                             policy->family,
1214                                             policy->selector.prefixlen_d, dir);
1215         write_seqcount_end(&bin->count);
1216         if (!n)
1217                 return NULL;
1218
1219         /* saddr is wildcard */
1220         if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr,
1221                                                policy->family,
1222                                                policy->selector.prefixlen_s))
1223                 return &n->hhead;
1224
1225         write_seqcount_begin(&bin->count);
1226         n = xfrm_policy_inexact_insert_node(net,
1227                                             &n->root,
1228                                             &policy->selector.saddr,
1229                                             policy->family,
1230                                             policy->selector.prefixlen_s, dir);
1231         write_seqcount_end(&bin->count);
1232         if (!n)
1233                 return NULL;
1234
1235         return &n->hhead;
1236 }
1237
1238 static struct xfrm_policy *
1239 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl)
1240 {
1241         struct xfrm_pol_inexact_bin *bin;
1242         struct xfrm_policy *delpol;
1243         struct hlist_head *chain;
1244         struct net *net;
1245
1246         bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1247         if (!bin)
1248                 return ERR_PTR(-ENOMEM);
1249
1250         net = xp_net(policy);
1251         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
1252
1253         chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir);
1254         if (!chain) {
1255                 __xfrm_policy_inexact_prune_bin(bin, false);
1256                 return ERR_PTR(-ENOMEM);
1257         }
1258
1259         delpol = xfrm_policy_insert_list(chain, policy, excl);
1260         if (delpol && excl) {
1261                 __xfrm_policy_inexact_prune_bin(bin, false);
1262                 return ERR_PTR(-EEXIST);
1263         }
1264
1265         if (delpol)
1266                 __xfrm_policy_inexact_prune_bin(bin, false);
1267
1268         return delpol;
1269 }
1270
1271 static bool xfrm_policy_is_dead_or_sk(const struct xfrm_policy *policy)
1272 {
1273         int dir;
1274
1275         if (policy->walk.dead)
1276                 return true;
1277
1278         dir = xfrm_policy_id2dir(policy->index);
1279         return dir >= XFRM_POLICY_MAX;
1280 }
1281
1282 static void xfrm_hash_rebuild(struct work_struct *work)
1283 {
1284         struct net *net = container_of(work, struct net,
1285                                        xfrm.policy_hthresh.work);
1286         struct xfrm_policy *pol;
1287         struct xfrm_policy *policy;
1288         struct hlist_head *chain;
1289         struct hlist_node *newpos;
1290         int dir;
1291         unsigned seq;
1292         u8 lbits4, rbits4, lbits6, rbits6;
1293
1294         mutex_lock(&hash_resize_mutex);
1295
1296         /* read selector prefixlen thresholds */
1297         do {
1298                 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock);
1299
1300                 lbits4 = net->xfrm.policy_hthresh.lbits4;
1301                 rbits4 = net->xfrm.policy_hthresh.rbits4;
1302                 lbits6 = net->xfrm.policy_hthresh.lbits6;
1303                 rbits6 = net->xfrm.policy_hthresh.rbits6;
1304         } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq));
1305
1306         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1307         write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
1308
1309         /* make sure that we can insert the indirect policies again before
1310          * we start with destructive action.
1311          */
1312         list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
1313                 struct xfrm_pol_inexact_bin *bin;
1314                 u8 dbits, sbits;
1315
1316                 if (xfrm_policy_is_dead_or_sk(policy))
1317                         continue;
1318
1319                 dir = xfrm_policy_id2dir(policy->index);
1320                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1321                         if (policy->family == AF_INET) {
1322                                 dbits = rbits4;
1323                                 sbits = lbits4;
1324                         } else {
1325                                 dbits = rbits6;
1326                                 sbits = lbits6;
1327                         }
1328                 } else {
1329                         if (policy->family == AF_INET) {
1330                                 dbits = lbits4;
1331                                 sbits = rbits4;
1332                         } else {
1333                                 dbits = lbits6;
1334                                 sbits = rbits6;
1335                         }
1336                 }
1337
1338                 if (policy->selector.prefixlen_d < dbits ||
1339                     policy->selector.prefixlen_s < sbits)
1340                         continue;
1341
1342                 bin = xfrm_policy_inexact_alloc_bin(policy, dir);
1343                 if (!bin)
1344                         goto out_unlock;
1345
1346                 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir))
1347                         goto out_unlock;
1348         }
1349
1350         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
1351                 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) {
1352                         /* dir out => dst = remote, src = local */
1353                         net->xfrm.policy_bydst[dir].dbits4 = rbits4;
1354                         net->xfrm.policy_bydst[dir].sbits4 = lbits4;
1355                         net->xfrm.policy_bydst[dir].dbits6 = rbits6;
1356                         net->xfrm.policy_bydst[dir].sbits6 = lbits6;
1357                 } else {
1358                         /* dir in/fwd => dst = local, src = remote */
1359                         net->xfrm.policy_bydst[dir].dbits4 = lbits4;
1360                         net->xfrm.policy_bydst[dir].sbits4 = rbits4;
1361                         net->xfrm.policy_bydst[dir].dbits6 = lbits6;
1362                         net->xfrm.policy_bydst[dir].sbits6 = rbits6;
1363                 }
1364         }
1365
1366         /* re-insert all policies by order of creation */
1367         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
1368                 if (xfrm_policy_is_dead_or_sk(policy))
1369                         continue;
1370
1371                 hlist_del_rcu(&policy->bydst);
1372
1373                 newpos = NULL;
1374                 dir = xfrm_policy_id2dir(policy->index);
1375                 chain = policy_hash_bysel(net, &policy->selector,
1376                                           policy->family, dir);
1377
1378                 if (!chain) {
1379                         void *p = xfrm_policy_inexact_insert(policy, dir, 0);
1380
1381                         WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p));
1382                         continue;
1383                 }
1384
1385                 hlist_for_each_entry(pol, chain, bydst) {
1386                         if (policy->priority >= pol->priority)
1387                                 newpos = &pol->bydst;
1388                         else
1389                                 break;
1390                 }
1391                 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1392                         hlist_add_behind_rcu(&policy->bydst, newpos);
1393                 else
1394                         hlist_add_head_rcu(&policy->bydst, chain);
1395         }
1396
1397 out_unlock:
1398         __xfrm_policy_inexact_flush(net);
1399         write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation);
1400         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1401
1402         mutex_unlock(&hash_resize_mutex);
1403 }
1404
1405 void xfrm_policy_hash_rebuild(struct net *net)
1406 {
1407         schedule_work(&net->xfrm.policy_hthresh.work);
1408 }
1409 EXPORT_SYMBOL(xfrm_policy_hash_rebuild);
1410
1411 /* Generate new index... KAME seems to generate them ordered by cost
1412  * of an absolute inpredictability of ordering of rules. This will not pass. */
1413 static u32 xfrm_gen_index(struct net *net, int dir, u32 index)
1414 {
1415         for (;;) {
1416                 struct hlist_head *list;
1417                 struct xfrm_policy *p;
1418                 u32 idx;
1419                 int found;
1420
1421                 if (!index) {
1422                         idx = (net->xfrm.idx_generator | dir);
1423                         net->xfrm.idx_generator += 8;
1424                 } else {
1425                         idx = index;
1426                         index = 0;
1427                 }
1428
1429                 if (idx == 0)
1430                         idx = 8;
1431                 list = net->xfrm.policy_byidx + idx_hash(net, idx);
1432                 found = 0;
1433                 hlist_for_each_entry(p, list, byidx) {
1434                         if (p->index == idx) {
1435                                 found = 1;
1436                                 break;
1437                         }
1438                 }
1439                 if (!found)
1440                         return idx;
1441         }
1442 }
1443
1444 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
1445 {
1446         u32 *p1 = (u32 *) s1;
1447         u32 *p2 = (u32 *) s2;
1448         int len = sizeof(struct xfrm_selector) / sizeof(u32);
1449         int i;
1450
1451         for (i = 0; i < len; i++) {
1452                 if (p1[i] != p2[i])
1453                         return 1;
1454         }
1455
1456         return 0;
1457 }
1458
1459 static void xfrm_policy_requeue(struct xfrm_policy *old,
1460                                 struct xfrm_policy *new)
1461 {
1462         struct xfrm_policy_queue *pq = &old->polq;
1463         struct sk_buff_head list;
1464
1465         if (skb_queue_empty(&pq->hold_queue))
1466                 return;
1467
1468         __skb_queue_head_init(&list);
1469
1470         spin_lock_bh(&pq->hold_queue.lock);
1471         skb_queue_splice_init(&pq->hold_queue, &list);
1472         if (del_timer(&pq->hold_timer))
1473                 xfrm_pol_put(old);
1474         spin_unlock_bh(&pq->hold_queue.lock);
1475
1476         pq = &new->polq;
1477
1478         spin_lock_bh(&pq->hold_queue.lock);
1479         skb_queue_splice(&list, &pq->hold_queue);
1480         pq->timeout = XFRM_QUEUE_TMO_MIN;
1481         if (!mod_timer(&pq->hold_timer, jiffies))
1482                 xfrm_pol_hold(new);
1483         spin_unlock_bh(&pq->hold_queue.lock);
1484 }
1485
1486 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark,
1487                                           struct xfrm_policy *pol)
1488 {
1489         return mark->v == pol->mark.v && mark->m == pol->mark.m;
1490 }
1491
1492 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed)
1493 {
1494         const struct xfrm_pol_inexact_key *k = data;
1495         u32 a = k->type << 24 | k->dir << 16 | k->family;
1496
1497         return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)),
1498                             seed);
1499 }
1500
1501 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed)
1502 {
1503         const struct xfrm_pol_inexact_bin *b = data;
1504
1505         return xfrm_pol_bin_key(&b->k, 0, seed);
1506 }
1507
1508 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg,
1509                             const void *ptr)
1510 {
1511         const struct xfrm_pol_inexact_key *key = arg->key;
1512         const struct xfrm_pol_inexact_bin *b = ptr;
1513         int ret;
1514
1515         if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net)))
1516                 return -1;
1517
1518         ret = b->k.dir ^ key->dir;
1519         if (ret)
1520                 return ret;
1521
1522         ret = b->k.type ^ key->type;
1523         if (ret)
1524                 return ret;
1525
1526         ret = b->k.family ^ key->family;
1527         if (ret)
1528                 return ret;
1529
1530         return b->k.if_id ^ key->if_id;
1531 }
1532
1533 static const struct rhashtable_params xfrm_pol_inexact_params = {
1534         .head_offset            = offsetof(struct xfrm_pol_inexact_bin, head),
1535         .hashfn                 = xfrm_pol_bin_key,
1536         .obj_hashfn             = xfrm_pol_bin_obj,
1537         .obj_cmpfn              = xfrm_pol_bin_cmp,
1538         .automatic_shrinking    = true,
1539 };
1540
1541 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain,
1542                                                    struct xfrm_policy *policy,
1543                                                    bool excl)
1544 {
1545         struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL;
1546
1547         hlist_for_each_entry(pol, chain, bydst) {
1548                 if (pol->type == policy->type &&
1549                     pol->if_id == policy->if_id &&
1550                     !selector_cmp(&pol->selector, &policy->selector) &&
1551                     xfrm_policy_mark_match(&policy->mark, pol) &&
1552                     xfrm_sec_ctx_match(pol->security, policy->security) &&
1553                     !WARN_ON(delpol)) {
1554                         if (excl)
1555                                 return ERR_PTR(-EEXIST);
1556                         delpol = pol;
1557                         if (policy->priority > pol->priority)
1558                                 continue;
1559                 } else if (policy->priority >= pol->priority) {
1560                         newpos = pol;
1561                         continue;
1562                 }
1563                 if (delpol)
1564                         break;
1565         }
1566
1567         if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET)
1568                 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst);
1569         else
1570                 /* Packet offload policies enter to the head
1571                  * to speed-up lookups.
1572                  */
1573                 hlist_add_head_rcu(&policy->bydst, chain);
1574
1575         return delpol;
1576 }
1577
1578 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
1579 {
1580         struct net *net = xp_net(policy);
1581         struct xfrm_policy *delpol;
1582         struct hlist_head *chain;
1583
1584         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1585         chain = policy_hash_bysel(net, &policy->selector, policy->family, dir);
1586         if (chain)
1587                 delpol = xfrm_policy_insert_list(chain, policy, excl);
1588         else
1589                 delpol = xfrm_policy_inexact_insert(policy, dir, excl);
1590
1591         if (IS_ERR(delpol)) {
1592                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1593                 return PTR_ERR(delpol);
1594         }
1595
1596         __xfrm_policy_link(policy, dir);
1597
1598         /* After previous checking, family can either be AF_INET or AF_INET6 */
1599         if (policy->family == AF_INET)
1600                 rt_genid_bump_ipv4(net);
1601         else
1602                 rt_genid_bump_ipv6(net);
1603
1604         if (delpol) {
1605                 xfrm_policy_requeue(delpol, policy);
1606                 __xfrm_policy_unlink(delpol, dir);
1607         }
1608         policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index);
1609         hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
1610         policy->curlft.add_time = ktime_get_real_seconds();
1611         policy->curlft.use_time = 0;
1612         if (!mod_timer(&policy->timer, jiffies + HZ))
1613                 xfrm_pol_hold(policy);
1614         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1615
1616         if (delpol)
1617                 xfrm_policy_kill(delpol);
1618         else if (xfrm_bydst_should_resize(net, dir, NULL))
1619                 schedule_work(&net->xfrm.policy_hash_work);
1620
1621         return 0;
1622 }
1623 EXPORT_SYMBOL(xfrm_policy_insert);
1624
1625 static struct xfrm_policy *
1626 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark,
1627                         u32 if_id, u8 type, int dir, struct xfrm_selector *sel,
1628                         struct xfrm_sec_ctx *ctx)
1629 {
1630         struct xfrm_policy *pol;
1631
1632         if (!chain)
1633                 return NULL;
1634
1635         hlist_for_each_entry(pol, chain, bydst) {
1636                 if (pol->type == type &&
1637                     pol->if_id == if_id &&
1638                     xfrm_policy_mark_match(mark, pol) &&
1639                     !selector_cmp(sel, &pol->selector) &&
1640                     xfrm_sec_ctx_match(ctx, pol->security))
1641                         return pol;
1642         }
1643
1644         return NULL;
1645 }
1646
1647 struct xfrm_policy *
1648 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1649                       u8 type, int dir, struct xfrm_selector *sel,
1650                       struct xfrm_sec_ctx *ctx, int delete, int *err)
1651 {
1652         struct xfrm_pol_inexact_bin *bin = NULL;
1653         struct xfrm_policy *pol, *ret = NULL;
1654         struct hlist_head *chain;
1655
1656         *err = 0;
1657         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1658         chain = policy_hash_bysel(net, sel, sel->family, dir);
1659         if (!chain) {
1660                 struct xfrm_pol_inexact_candidates cand;
1661                 int i;
1662
1663                 bin = xfrm_policy_inexact_lookup(net, type,
1664                                                  sel->family, dir, if_id);
1665                 if (!bin) {
1666                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1667                         return NULL;
1668                 }
1669
1670                 if (!xfrm_policy_find_inexact_candidates(&cand, bin,
1671                                                          &sel->saddr,
1672                                                          &sel->daddr)) {
1673                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1674                         return NULL;
1675                 }
1676
1677                 pol = NULL;
1678                 for (i = 0; i < ARRAY_SIZE(cand.res); i++) {
1679                         struct xfrm_policy *tmp;
1680
1681                         tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark,
1682                                                       if_id, type, dir,
1683                                                       sel, ctx);
1684                         if (!tmp)
1685                                 continue;
1686
1687                         if (!pol || tmp->pos < pol->pos)
1688                                 pol = tmp;
1689                 }
1690         } else {
1691                 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir,
1692                                               sel, ctx);
1693         }
1694
1695         if (pol) {
1696                 xfrm_pol_hold(pol);
1697                 if (delete) {
1698                         *err = security_xfrm_policy_delete(pol->security);
1699                         if (*err) {
1700                                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1701                                 return pol;
1702                         }
1703                         __xfrm_policy_unlink(pol, dir);
1704                 }
1705                 ret = pol;
1706         }
1707         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1708
1709         if (ret && delete)
1710                 xfrm_policy_kill(ret);
1711         if (bin && delete)
1712                 xfrm_policy_inexact_prune_bin(bin);
1713         return ret;
1714 }
1715 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1716
1717 struct xfrm_policy *
1718 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id,
1719                  u8 type, int dir, u32 id, int delete, int *err)
1720 {
1721         struct xfrm_policy *pol, *ret;
1722         struct hlist_head *chain;
1723
1724         *err = -ENOENT;
1725         if (xfrm_policy_id2dir(id) != dir)
1726                 return NULL;
1727
1728         *err = 0;
1729         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1730         chain = net->xfrm.policy_byidx + idx_hash(net, id);
1731         ret = NULL;
1732         hlist_for_each_entry(pol, chain, byidx) {
1733                 if (pol->type == type && pol->index == id &&
1734                     pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) {
1735                         xfrm_pol_hold(pol);
1736                         if (delete) {
1737                                 *err = security_xfrm_policy_delete(
1738                                                                 pol->security);
1739                                 if (*err) {
1740                                         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1741                                         return pol;
1742                                 }
1743                                 __xfrm_policy_unlink(pol, dir);
1744                         }
1745                         ret = pol;
1746                         break;
1747                 }
1748         }
1749         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1750
1751         if (ret && delete)
1752                 xfrm_policy_kill(ret);
1753         return ret;
1754 }
1755 EXPORT_SYMBOL(xfrm_policy_byid);
1756
1757 #ifdef CONFIG_SECURITY_NETWORK_XFRM
1758 static inline int
1759 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1760 {
1761         struct xfrm_policy *pol;
1762         int err = 0;
1763
1764         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1765                 if (pol->walk.dead ||
1766                     xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1767                     pol->type != type)
1768                         continue;
1769
1770                 err = security_xfrm_policy_delete(pol->security);
1771                 if (err) {
1772                         xfrm_audit_policy_delete(pol, 0, task_valid);
1773                         return err;
1774                 }
1775         }
1776         return err;
1777 }
1778
1779 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1780                                                      struct net_device *dev,
1781                                                      bool task_valid)
1782 {
1783         struct xfrm_policy *pol;
1784         int err = 0;
1785
1786         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1787                 if (pol->walk.dead ||
1788                     xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX ||
1789                     pol->xdo.dev != dev)
1790                         continue;
1791
1792                 err = security_xfrm_policy_delete(pol->security);
1793                 if (err) {
1794                         xfrm_audit_policy_delete(pol, 0, task_valid);
1795                         return err;
1796                 }
1797         }
1798         return err;
1799 }
1800 #else
1801 static inline int
1802 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid)
1803 {
1804         return 0;
1805 }
1806
1807 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net,
1808                                                      struct net_device *dev,
1809                                                      bool task_valid)
1810 {
1811         return 0;
1812 }
1813 #endif
1814
1815 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid)
1816 {
1817         int dir, err = 0, cnt = 0;
1818         struct xfrm_policy *pol;
1819
1820         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1821
1822         err = xfrm_policy_flush_secctx_check(net, type, task_valid);
1823         if (err)
1824                 goto out;
1825
1826 again:
1827         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1828                 if (pol->walk.dead)
1829                         continue;
1830
1831                 dir = xfrm_policy_id2dir(pol->index);
1832                 if (dir >= XFRM_POLICY_MAX ||
1833                     pol->type != type)
1834                         continue;
1835
1836                 __xfrm_policy_unlink(pol, dir);
1837                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1838                 cnt++;
1839                 xfrm_audit_policy_delete(pol, 1, task_valid);
1840                 xfrm_policy_kill(pol);
1841                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1842                 goto again;
1843         }
1844         if (cnt)
1845                 __xfrm_policy_inexact_flush(net);
1846         else
1847                 err = -ESRCH;
1848 out:
1849         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1850         return err;
1851 }
1852 EXPORT_SYMBOL(xfrm_policy_flush);
1853
1854 int xfrm_dev_policy_flush(struct net *net, struct net_device *dev,
1855                           bool task_valid)
1856 {
1857         int dir, err = 0, cnt = 0;
1858         struct xfrm_policy *pol;
1859
1860         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1861
1862         err = xfrm_dev_policy_flush_secctx_check(net, dev, task_valid);
1863         if (err)
1864                 goto out;
1865
1866 again:
1867         list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) {
1868                 if (pol->walk.dead)
1869                         continue;
1870
1871                 dir = xfrm_policy_id2dir(pol->index);
1872                 if (dir >= XFRM_POLICY_MAX ||
1873                     pol->xdo.dev != dev)
1874                         continue;
1875
1876                 __xfrm_policy_unlink(pol, dir);
1877                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1878                 cnt++;
1879                 xfrm_audit_policy_delete(pol, 1, task_valid);
1880                 xfrm_policy_kill(pol);
1881                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1882                 goto again;
1883         }
1884         if (cnt)
1885                 __xfrm_policy_inexact_flush(net);
1886         else
1887                 err = -ESRCH;
1888 out:
1889         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1890         return err;
1891 }
1892 EXPORT_SYMBOL(xfrm_dev_policy_flush);
1893
1894 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
1895                      int (*func)(struct xfrm_policy *, int, int, void*),
1896                      void *data)
1897 {
1898         struct xfrm_policy *pol;
1899         struct xfrm_policy_walk_entry *x;
1900         int error = 0;
1901
1902         if (walk->type >= XFRM_POLICY_TYPE_MAX &&
1903             walk->type != XFRM_POLICY_TYPE_ANY)
1904                 return -EINVAL;
1905
1906         if (list_empty(&walk->walk.all) && walk->seq != 0)
1907                 return 0;
1908
1909         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
1910         if (list_empty(&walk->walk.all))
1911                 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
1912         else
1913                 x = list_first_entry(&walk->walk.all,
1914                                      struct xfrm_policy_walk_entry, all);
1915
1916         list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
1917                 if (x->dead)
1918                         continue;
1919                 pol = container_of(x, struct xfrm_policy, walk);
1920                 if (walk->type != XFRM_POLICY_TYPE_ANY &&
1921                     walk->type != pol->type)
1922                         continue;
1923                 error = func(pol, xfrm_policy_id2dir(pol->index),
1924                              walk->seq, data);
1925                 if (error) {
1926                         list_move_tail(&walk->walk.all, &x->all);
1927                         goto out;
1928                 }
1929                 walk->seq++;
1930         }
1931         if (walk->seq == 0) {
1932                 error = -ENOENT;
1933                 goto out;
1934         }
1935         list_del_init(&walk->walk.all);
1936 out:
1937         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1938         return error;
1939 }
1940 EXPORT_SYMBOL(xfrm_policy_walk);
1941
1942 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type)
1943 {
1944         INIT_LIST_HEAD(&walk->walk.all);
1945         walk->walk.dead = 1;
1946         walk->type = type;
1947         walk->seq = 0;
1948 }
1949 EXPORT_SYMBOL(xfrm_policy_walk_init);
1950
1951 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net)
1952 {
1953         if (list_empty(&walk->walk.all))
1954                 return;
1955
1956         spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */
1957         list_del(&walk->walk.all);
1958         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
1959 }
1960 EXPORT_SYMBOL(xfrm_policy_walk_done);
1961
1962 /*
1963  * Find policy to apply to this flow.
1964  *
1965  * Returns 0 if policy found, else an -errno.
1966  */
1967 static int xfrm_policy_match(const struct xfrm_policy *pol,
1968                              const struct flowi *fl,
1969                              u8 type, u16 family, u32 if_id)
1970 {
1971         const struct xfrm_selector *sel = &pol->selector;
1972         int ret = -ESRCH;
1973         bool match;
1974
1975         if (pol->family != family ||
1976             pol->if_id != if_id ||
1977             (fl->flowi_mark & pol->mark.m) != pol->mark.v ||
1978             pol->type != type)
1979                 return ret;
1980
1981         match = xfrm_selector_match(sel, fl, family);
1982         if (match)
1983                 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid);
1984         return ret;
1985 }
1986
1987 static struct xfrm_pol_inexact_node *
1988 xfrm_policy_lookup_inexact_addr(const struct rb_root *r,
1989                                 seqcount_spinlock_t *count,
1990                                 const xfrm_address_t *addr, u16 family)
1991 {
1992         const struct rb_node *parent;
1993         int seq;
1994
1995 again:
1996         seq = read_seqcount_begin(count);
1997
1998         parent = rcu_dereference_raw(r->rb_node);
1999         while (parent) {
2000                 struct xfrm_pol_inexact_node *node;
2001                 int delta;
2002
2003                 node = rb_entry(parent, struct xfrm_pol_inexact_node, node);
2004
2005                 delta = xfrm_policy_addr_delta(addr, &node->addr,
2006                                                node->prefixlen, family);
2007                 if (delta < 0) {
2008                         parent = rcu_dereference_raw(parent->rb_left);
2009                         continue;
2010                 } else if (delta > 0) {
2011                         parent = rcu_dereference_raw(parent->rb_right);
2012                         continue;
2013                 }
2014
2015                 return node;
2016         }
2017
2018         if (read_seqcount_retry(count, seq))
2019                 goto again;
2020
2021         return NULL;
2022 }
2023
2024 static bool
2025 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand,
2026                                     struct xfrm_pol_inexact_bin *b,
2027                                     const xfrm_address_t *saddr,
2028                                     const xfrm_address_t *daddr)
2029 {
2030         struct xfrm_pol_inexact_node *n;
2031         u16 family;
2032
2033         if (!b)
2034                 return false;
2035
2036         family = b->k.family;
2037         memset(cand, 0, sizeof(*cand));
2038         cand->res[XFRM_POL_CAND_ANY] = &b->hhead;
2039
2040         n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr,
2041                                             family);
2042         if (n) {
2043                 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead;
2044                 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr,
2045                                                     family);
2046                 if (n)
2047                         cand->res[XFRM_POL_CAND_BOTH] = &n->hhead;
2048         }
2049
2050         n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr,
2051                                             family);
2052         if (n)
2053                 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead;
2054
2055         return true;
2056 }
2057
2058 static struct xfrm_pol_inexact_bin *
2059 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family,
2060                                u8 dir, u32 if_id)
2061 {
2062         struct xfrm_pol_inexact_key k = {
2063                 .family = family,
2064                 .type = type,
2065                 .dir = dir,
2066                 .if_id = if_id,
2067         };
2068
2069         write_pnet(&k.net, net);
2070
2071         return rhashtable_lookup(&xfrm_policy_inexact_table, &k,
2072                                  xfrm_pol_inexact_params);
2073 }
2074
2075 static struct xfrm_pol_inexact_bin *
2076 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family,
2077                            u8 dir, u32 if_id)
2078 {
2079         struct xfrm_pol_inexact_bin *bin;
2080
2081         lockdep_assert_held(&net->xfrm.xfrm_policy_lock);
2082
2083         rcu_read_lock();
2084         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2085         rcu_read_unlock();
2086
2087         return bin;
2088 }
2089
2090 static struct xfrm_policy *
2091 __xfrm_policy_eval_candidates(struct hlist_head *chain,
2092                               struct xfrm_policy *prefer,
2093                               const struct flowi *fl,
2094                               u8 type, u16 family, u32 if_id)
2095 {
2096         u32 priority = prefer ? prefer->priority : ~0u;
2097         struct xfrm_policy *pol;
2098
2099         if (!chain)
2100                 return NULL;
2101
2102         hlist_for_each_entry_rcu(pol, chain, bydst) {
2103                 int err;
2104
2105                 if (pol->priority > priority)
2106                         break;
2107
2108                 err = xfrm_policy_match(pol, fl, type, family, if_id);
2109                 if (err) {
2110                         if (err != -ESRCH)
2111                                 return ERR_PTR(err);
2112
2113                         continue;
2114                 }
2115
2116                 if (prefer) {
2117                         /* matches.  Is it older than *prefer? */
2118                         if (pol->priority == priority &&
2119                             prefer->pos < pol->pos)
2120                                 return prefer;
2121                 }
2122
2123                 return pol;
2124         }
2125
2126         return NULL;
2127 }
2128
2129 static struct xfrm_policy *
2130 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand,
2131                             struct xfrm_policy *prefer,
2132                             const struct flowi *fl,
2133                             u8 type, u16 family, u32 if_id)
2134 {
2135         struct xfrm_policy *tmp;
2136         int i;
2137
2138         for (i = 0; i < ARRAY_SIZE(cand->res); i++) {
2139                 tmp = __xfrm_policy_eval_candidates(cand->res[i],
2140                                                     prefer,
2141                                                     fl, type, family, if_id);
2142                 if (!tmp)
2143                         continue;
2144
2145                 if (IS_ERR(tmp))
2146                         return tmp;
2147                 prefer = tmp;
2148         }
2149
2150         return prefer;
2151 }
2152
2153 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
2154                                                      const struct flowi *fl,
2155                                                      u16 family, u8 dir,
2156                                                      u32 if_id)
2157 {
2158         struct xfrm_pol_inexact_candidates cand;
2159         const xfrm_address_t *daddr, *saddr;
2160         struct xfrm_pol_inexact_bin *bin;
2161         struct xfrm_policy *pol, *ret;
2162         struct hlist_head *chain;
2163         unsigned int sequence;
2164         int err;
2165
2166         daddr = xfrm_flowi_daddr(fl, family);
2167         saddr = xfrm_flowi_saddr(fl, family);
2168         if (unlikely(!daddr || !saddr))
2169                 return NULL;
2170
2171         rcu_read_lock();
2172  retry:
2173         do {
2174                 sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation);
2175                 chain = policy_hash_direct(net, daddr, saddr, family, dir);
2176         } while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence));
2177
2178         ret = NULL;
2179         hlist_for_each_entry_rcu(pol, chain, bydst) {
2180                 err = xfrm_policy_match(pol, fl, type, family, if_id);
2181                 if (err) {
2182                         if (err == -ESRCH)
2183                                 continue;
2184                         else {
2185                                 ret = ERR_PTR(err);
2186                                 goto fail;
2187                         }
2188                 } else {
2189                         ret = pol;
2190                         break;
2191                 }
2192         }
2193         if (ret && ret->xdo.type == XFRM_DEV_OFFLOAD_PACKET)
2194                 goto skip_inexact;
2195
2196         bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id);
2197         if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr,
2198                                                          daddr))
2199                 goto skip_inexact;
2200
2201         pol = xfrm_policy_eval_candidates(&cand, ret, fl, type,
2202                                           family, if_id);
2203         if (pol) {
2204                 ret = pol;
2205                 if (IS_ERR(pol))
2206                         goto fail;
2207         }
2208
2209 skip_inexact:
2210         if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence))
2211                 goto retry;
2212
2213         if (ret && !xfrm_pol_hold_rcu(ret))
2214                 goto retry;
2215 fail:
2216         rcu_read_unlock();
2217
2218         return ret;
2219 }
2220
2221 static struct xfrm_policy *xfrm_policy_lookup(struct net *net,
2222                                               const struct flowi *fl,
2223                                               u16 family, u8 dir, u32 if_id)
2224 {
2225 #ifdef CONFIG_XFRM_SUB_POLICY
2226         struct xfrm_policy *pol;
2227
2228         pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family,
2229                                         dir, if_id);
2230         if (pol != NULL)
2231                 return pol;
2232 #endif
2233         return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family,
2234                                          dir, if_id);
2235 }
2236
2237 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir,
2238                                                  const struct flowi *fl,
2239                                                  u16 family, u32 if_id)
2240 {
2241         struct xfrm_policy *pol;
2242
2243         rcu_read_lock();
2244  again:
2245         pol = rcu_dereference(sk->sk_policy[dir]);
2246         if (pol != NULL) {
2247                 bool match;
2248                 int err = 0;
2249
2250                 if (pol->family != family) {
2251                         pol = NULL;
2252                         goto out;
2253                 }
2254
2255                 match = xfrm_selector_match(&pol->selector, fl, family);
2256                 if (match) {
2257                         if ((READ_ONCE(sk->sk_mark) & pol->mark.m) != pol->mark.v ||
2258                             pol->if_id != if_id) {
2259                                 pol = NULL;
2260                                 goto out;
2261                         }
2262                         err = security_xfrm_policy_lookup(pol->security,
2263                                                       fl->flowi_secid);
2264                         if (!err) {
2265                                 if (!xfrm_pol_hold_rcu(pol))
2266                                         goto again;
2267                         } else if (err == -ESRCH) {
2268                                 pol = NULL;
2269                         } else {
2270                                 pol = ERR_PTR(err);
2271                         }
2272                 } else
2273                         pol = NULL;
2274         }
2275 out:
2276         rcu_read_unlock();
2277         return pol;
2278 }
2279
2280 static u32 xfrm_gen_pos_slow(struct net *net)
2281 {
2282         struct xfrm_policy *policy;
2283         u32 i = 0;
2284
2285         /* oldest entry is last in list */
2286         list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) {
2287                 if (!xfrm_policy_is_dead_or_sk(policy))
2288                         policy->pos = ++i;
2289         }
2290
2291         return i;
2292 }
2293
2294 static u32 xfrm_gen_pos(struct net *net)
2295 {
2296         const struct xfrm_policy *policy;
2297         u32 i = 0;
2298
2299         /* most recently added policy is at the head of the list */
2300         list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) {
2301                 if (xfrm_policy_is_dead_or_sk(policy))
2302                         continue;
2303
2304                 if (policy->pos == UINT_MAX)
2305                         return xfrm_gen_pos_slow(net);
2306
2307                 i = policy->pos + 1;
2308                 break;
2309         }
2310
2311         return i;
2312 }
2313
2314 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
2315 {
2316         struct net *net = xp_net(pol);
2317
2318         switch (dir) {
2319         case XFRM_POLICY_IN:
2320         case XFRM_POLICY_FWD:
2321         case XFRM_POLICY_OUT:
2322                 pol->pos = xfrm_gen_pos(net);
2323                 break;
2324         }
2325
2326         list_add(&pol->walk.all, &net->xfrm.policy_all);
2327         net->xfrm.policy_count[dir]++;
2328         xfrm_pol_hold(pol);
2329 }
2330
2331 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
2332                                                 int dir)
2333 {
2334         struct net *net = xp_net(pol);
2335
2336         if (list_empty(&pol->walk.all))
2337                 return NULL;
2338
2339         /* Socket policies are not hashed. */
2340         if (!hlist_unhashed(&pol->bydst)) {
2341                 hlist_del_rcu(&pol->bydst);
2342                 hlist_del(&pol->byidx);
2343         }
2344
2345         list_del_init(&pol->walk.all);
2346         net->xfrm.policy_count[dir]--;
2347
2348         return pol;
2349 }
2350
2351 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir)
2352 {
2353         __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir);
2354 }
2355
2356 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir)
2357 {
2358         __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir);
2359 }
2360
2361 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
2362 {
2363         struct net *net = xp_net(pol);
2364
2365         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2366         pol = __xfrm_policy_unlink(pol, dir);
2367         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2368         if (pol) {
2369                 xfrm_policy_kill(pol);
2370                 return 0;
2371         }
2372         return -ENOENT;
2373 }
2374 EXPORT_SYMBOL(xfrm_policy_delete);
2375
2376 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
2377 {
2378         struct net *net = sock_net(sk);
2379         struct xfrm_policy *old_pol;
2380
2381 #ifdef CONFIG_XFRM_SUB_POLICY
2382         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
2383                 return -EINVAL;
2384 #endif
2385
2386         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2387         old_pol = rcu_dereference_protected(sk->sk_policy[dir],
2388                                 lockdep_is_held(&net->xfrm.xfrm_policy_lock));
2389         if (pol) {
2390                 pol->curlft.add_time = ktime_get_real_seconds();
2391                 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0);
2392                 xfrm_sk_policy_link(pol, dir);
2393         }
2394         rcu_assign_pointer(sk->sk_policy[dir], pol);
2395         if (old_pol) {
2396                 if (pol)
2397                         xfrm_policy_requeue(old_pol, pol);
2398
2399                 /* Unlinking succeeds always. This is the only function
2400                  * allowed to delete or replace socket policy.
2401                  */
2402                 xfrm_sk_policy_unlink(old_pol, dir);
2403         }
2404         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2405
2406         if (old_pol) {
2407                 xfrm_policy_kill(old_pol);
2408         }
2409         return 0;
2410 }
2411
2412 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir)
2413 {
2414         struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC);
2415         struct net *net = xp_net(old);
2416
2417         if (newp) {
2418                 newp->selector = old->selector;
2419                 if (security_xfrm_policy_clone(old->security,
2420                                                &newp->security)) {
2421                         kfree(newp);
2422                         return NULL;  /* ENOMEM */
2423                 }
2424                 newp->lft = old->lft;
2425                 newp->curlft = old->curlft;
2426                 newp->mark = old->mark;
2427                 newp->if_id = old->if_id;
2428                 newp->action = old->action;
2429                 newp->flags = old->flags;
2430                 newp->xfrm_nr = old->xfrm_nr;
2431                 newp->index = old->index;
2432                 newp->type = old->type;
2433                 newp->family = old->family;
2434                 memcpy(newp->xfrm_vec, old->xfrm_vec,
2435                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
2436                 spin_lock_bh(&net->xfrm.xfrm_policy_lock);
2437                 xfrm_sk_policy_link(newp, dir);
2438                 spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
2439                 xfrm_pol_put(newp);
2440         }
2441         return newp;
2442 }
2443
2444 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk)
2445 {
2446         const struct xfrm_policy *p;
2447         struct xfrm_policy *np;
2448         int i, ret = 0;
2449
2450         rcu_read_lock();
2451         for (i = 0; i < 2; i++) {
2452                 p = rcu_dereference(osk->sk_policy[i]);
2453                 if (p) {
2454                         np = clone_policy(p, i);
2455                         if (unlikely(!np)) {
2456                                 ret = -ENOMEM;
2457                                 break;
2458                         }
2459                         rcu_assign_pointer(sk->sk_policy[i], np);
2460                 }
2461         }
2462         rcu_read_unlock();
2463         return ret;
2464 }
2465
2466 static int
2467 xfrm_get_saddr(unsigned short family, xfrm_address_t *saddr,
2468                const struct xfrm_dst_lookup_params *params)
2469 {
2470         int err;
2471         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2472
2473         if (unlikely(afinfo == NULL))
2474                 return -EINVAL;
2475         err = afinfo->get_saddr(saddr, params);
2476         rcu_read_unlock();
2477         return err;
2478 }
2479
2480 /* Resolve list of templates for the flow, given policy. */
2481
2482 static int
2483 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl,
2484                       struct xfrm_state **xfrm, unsigned short family)
2485 {
2486         struct net *net = xp_net(policy);
2487         int nx;
2488         int i, error;
2489         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
2490         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
2491         xfrm_address_t tmp;
2492
2493         for (nx = 0, i = 0; i < policy->xfrm_nr; i++) {
2494                 struct xfrm_state *x;
2495                 xfrm_address_t *remote = daddr;
2496                 xfrm_address_t *local  = saddr;
2497                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
2498
2499                 if (tmpl->mode == XFRM_MODE_TUNNEL ||
2500                     tmpl->mode == XFRM_MODE_BEET) {
2501                         remote = &tmpl->id.daddr;
2502                         local = &tmpl->saddr;
2503                         if (xfrm_addr_any(local, tmpl->encap_family)) {
2504                                 struct xfrm_dst_lookup_params params;
2505
2506                                 memset(&params, 0, sizeof(params));
2507                                 params.net = net;
2508                                 params.oif = fl->flowi_oif;
2509                                 params.daddr = remote;
2510                                 error = xfrm_get_saddr(tmpl->encap_family, &tmp,
2511                                                        &params);
2512                                 if (error)
2513                                         goto fail;
2514                                 local = &tmp;
2515                         }
2516                 }
2517
2518                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error,
2519                                     family, policy->if_id);
2520                 if (x && x->dir && x->dir != XFRM_SA_DIR_OUT) {
2521                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEDIRERROR);
2522                         xfrm_state_put(x);
2523                         error = -EINVAL;
2524                         goto fail;
2525                 }
2526
2527                 if (x && x->km.state == XFRM_STATE_VALID) {
2528                         xfrm[nx++] = x;
2529                         daddr = remote;
2530                         saddr = local;
2531                         continue;
2532                 }
2533                 if (x) {
2534                         error = (x->km.state == XFRM_STATE_ERROR ?
2535                                  -EINVAL : -EAGAIN);
2536                         xfrm_state_put(x);
2537                 } else if (error == -ESRCH) {
2538                         error = -EAGAIN;
2539                 }
2540
2541                 if (!tmpl->optional)
2542                         goto fail;
2543         }
2544         return nx;
2545
2546 fail:
2547         for (nx--; nx >= 0; nx--)
2548                 xfrm_state_put(xfrm[nx]);
2549         return error;
2550 }
2551
2552 static int
2553 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl,
2554                   struct xfrm_state **xfrm, unsigned short family)
2555 {
2556         struct xfrm_state *tp[XFRM_MAX_DEPTH];
2557         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
2558         int cnx = 0;
2559         int error;
2560         int ret;
2561         int i;
2562
2563         for (i = 0; i < npols; i++) {
2564                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
2565                         error = -ENOBUFS;
2566                         goto fail;
2567                 }
2568
2569                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
2570                 if (ret < 0) {
2571                         error = ret;
2572                         goto fail;
2573                 } else
2574                         cnx += ret;
2575         }
2576
2577         /* found states are sorted for outbound processing */
2578         if (npols > 1)
2579                 xfrm_state_sort(xfrm, tpp, cnx, family);
2580
2581         return cnx;
2582
2583  fail:
2584         for (cnx--; cnx >= 0; cnx--)
2585                 xfrm_state_put(tpp[cnx]);
2586         return error;
2587
2588 }
2589
2590 static dscp_t xfrm_get_dscp(const struct flowi *fl, int family)
2591 {
2592         if (family == AF_INET)
2593                 return inet_dsfield_to_dscp(fl->u.ip4.flowi4_tos);
2594
2595         return 0;
2596 }
2597
2598 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family)
2599 {
2600         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
2601         struct dst_ops *dst_ops;
2602         struct xfrm_dst *xdst;
2603
2604         if (!afinfo)
2605                 return ERR_PTR(-EINVAL);
2606
2607         switch (family) {
2608         case AF_INET:
2609                 dst_ops = &net->xfrm.xfrm4_dst_ops;
2610                 break;
2611 #if IS_ENABLED(CONFIG_IPV6)
2612         case AF_INET6:
2613                 dst_ops = &net->xfrm.xfrm6_dst_ops;
2614                 break;
2615 #endif
2616         default:
2617                 BUG();
2618         }
2619         xdst = dst_alloc(dst_ops, NULL, DST_OBSOLETE_NONE, 0);
2620
2621         if (likely(xdst)) {
2622                 memset_after(xdst, 0, u.dst);
2623         } else
2624                 xdst = ERR_PTR(-ENOBUFS);
2625
2626         rcu_read_unlock();
2627
2628         return xdst;
2629 }
2630
2631 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst,
2632                            int nfheader_len)
2633 {
2634         if (dst->ops->family == AF_INET6) {
2635                 path->path_cookie = rt6_get_cookie(dst_rt6_info(dst));
2636                 path->u.rt6.rt6i_nfheader_len = nfheader_len;
2637         }
2638 }
2639
2640 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
2641                                 const struct flowi *fl)
2642 {
2643         const struct xfrm_policy_afinfo *afinfo =
2644                 xfrm_policy_get_afinfo(xdst->u.dst.ops->family);
2645         int err;
2646
2647         if (!afinfo)
2648                 return -EINVAL;
2649
2650         err = afinfo->fill_dst(xdst, dev, fl);
2651
2652         rcu_read_unlock();
2653
2654         return err;
2655 }
2656
2657
2658 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
2659  * all the metrics... Shortly, bundle a bundle.
2660  */
2661
2662 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
2663                                             struct xfrm_state **xfrm,
2664                                             struct xfrm_dst **bundle,
2665                                             int nx,
2666                                             const struct flowi *fl,
2667                                             struct dst_entry *dst)
2668 {
2669         const struct xfrm_state_afinfo *afinfo;
2670         const struct xfrm_mode *inner_mode;
2671         struct net *net = xp_net(policy);
2672         unsigned long now = jiffies;
2673         struct net_device *dev;
2674         struct xfrm_dst *xdst_prev = NULL;
2675         struct xfrm_dst *xdst0 = NULL;
2676         int i = 0;
2677         int err;
2678         int header_len = 0;
2679         int nfheader_len = 0;
2680         int trailer_len = 0;
2681         int family = policy->selector.family;
2682         xfrm_address_t saddr, daddr;
2683         dscp_t dscp;
2684
2685         xfrm_flowi_addr_get(fl, &saddr, &daddr, family);
2686
2687         dscp = xfrm_get_dscp(fl, family);
2688
2689         dst_hold(dst);
2690
2691         for (; i < nx; i++) {
2692                 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family);
2693                 struct dst_entry *dst1 = &xdst->u.dst;
2694
2695                 err = PTR_ERR(xdst);
2696                 if (IS_ERR(xdst)) {
2697                         dst_release(dst);
2698                         goto put_states;
2699                 }
2700
2701                 bundle[i] = xdst;
2702                 if (!xdst_prev)
2703                         xdst0 = xdst;
2704                 else
2705                         /* Ref count is taken during xfrm_alloc_dst()
2706                          * No need to do dst_clone() on dst1
2707                          */
2708                         xfrm_dst_set_child(xdst_prev, &xdst->u.dst);
2709
2710                 if (xfrm[i]->sel.family == AF_UNSPEC) {
2711                         inner_mode = xfrm_ip2inner_mode(xfrm[i],
2712                                                         xfrm_af2proto(family));
2713                         if (!inner_mode) {
2714                                 err = -EAFNOSUPPORT;
2715                                 dst_release(dst);
2716                                 goto put_states;
2717                         }
2718                 } else
2719                         inner_mode = &xfrm[i]->inner_mode;
2720
2721                 xdst->route = dst;
2722                 dst_copy_metrics(dst1, dst);
2723
2724                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
2725                         __u32 mark = 0;
2726                         int oif;
2727
2728                         if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m)
2729                                 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]);
2730
2731                         if (xfrm[i]->xso.type != XFRM_DEV_OFFLOAD_PACKET)
2732                                 family = xfrm[i]->props.family;
2733
2734                         oif = fl->flowi_oif ? : fl->flowi_l3mdev;
2735                         dst = xfrm_dst_lookup(xfrm[i], dscp, oif, &saddr,
2736                                               &daddr, family, mark);
2737                         err = PTR_ERR(dst);
2738                         if (IS_ERR(dst))
2739                                 goto put_states;
2740                 } else
2741                         dst_hold(dst);
2742
2743                 dst1->xfrm = xfrm[i];
2744                 xdst->xfrm_genid = xfrm[i]->genid;
2745
2746                 dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
2747                 dst1->lastuse = now;
2748
2749                 dst1->input = dst_discard;
2750
2751                 rcu_read_lock();
2752                 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family);
2753                 if (likely(afinfo))
2754                         dst1->output = afinfo->output;
2755                 else
2756                         dst1->output = dst_discard_out;
2757                 rcu_read_unlock();
2758
2759                 xdst_prev = xdst;
2760
2761                 header_len += xfrm[i]->props.header_len;
2762                 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT)
2763                         nfheader_len += xfrm[i]->props.header_len;
2764                 trailer_len += xfrm[i]->props.trailer_len;
2765         }
2766
2767         xfrm_dst_set_child(xdst_prev, dst);
2768         xdst0->path = dst;
2769
2770         err = -ENODEV;
2771         dev = dst->dev;
2772         if (!dev)
2773                 goto free_dst;
2774
2775         xfrm_init_path(xdst0, dst, nfheader_len);
2776         xfrm_init_pmtu(bundle, nx);
2777
2778         for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst;
2779              xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) {
2780                 err = xfrm_fill_dst(xdst_prev, dev, fl);
2781                 if (err)
2782                         goto free_dst;
2783
2784                 xdst_prev->u.dst.header_len = header_len;
2785                 xdst_prev->u.dst.trailer_len = trailer_len;
2786                 header_len -= xdst_prev->u.dst.xfrm->props.header_len;
2787                 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len;
2788         }
2789
2790         return &xdst0->u.dst;
2791
2792 put_states:
2793         for (; i < nx; i++)
2794                 xfrm_state_put(xfrm[i]);
2795 free_dst:
2796         if (xdst0)
2797                 dst_release_immediate(&xdst0->u.dst);
2798
2799         return ERR_PTR(err);
2800 }
2801
2802 static int xfrm_expand_policies(const struct flowi *fl, u16 family,
2803                                 struct xfrm_policy **pols,
2804                                 int *num_pols, int *num_xfrms)
2805 {
2806         int i;
2807
2808         if (*num_pols == 0 || !pols[0]) {
2809                 *num_pols = 0;
2810                 *num_xfrms = 0;
2811                 return 0;
2812         }
2813         if (IS_ERR(pols[0])) {
2814                 *num_pols = 0;
2815                 return PTR_ERR(pols[0]);
2816         }
2817
2818         *num_xfrms = pols[0]->xfrm_nr;
2819
2820 #ifdef CONFIG_XFRM_SUB_POLICY
2821         if (pols[0]->action == XFRM_POLICY_ALLOW &&
2822             pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
2823                 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]),
2824                                                     XFRM_POLICY_TYPE_MAIN,
2825                                                     fl, family,
2826                                                     XFRM_POLICY_OUT,
2827                                                     pols[0]->if_id);
2828                 if (pols[1]) {
2829                         if (IS_ERR(pols[1])) {
2830                                 xfrm_pols_put(pols, *num_pols);
2831                                 *num_pols = 0;
2832                                 return PTR_ERR(pols[1]);
2833                         }
2834                         (*num_pols)++;
2835                         (*num_xfrms) += pols[1]->xfrm_nr;
2836                 }
2837         }
2838 #endif
2839         for (i = 0; i < *num_pols; i++) {
2840                 if (pols[i]->action != XFRM_POLICY_ALLOW) {
2841                         *num_xfrms = -1;
2842                         break;
2843                 }
2844         }
2845
2846         return 0;
2847
2848 }
2849
2850 static struct xfrm_dst *
2851 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols,
2852                                const struct flowi *fl, u16 family,
2853                                struct dst_entry *dst_orig)
2854 {
2855         struct net *net = xp_net(pols[0]);
2856         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
2857         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
2858         struct xfrm_dst *xdst;
2859         struct dst_entry *dst;
2860         int err;
2861
2862         /* Try to instantiate a bundle */
2863         err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family);
2864         if (err <= 0) {
2865                 if (err == 0)
2866                         return NULL;
2867
2868                 if (err != -EAGAIN)
2869                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
2870                 return ERR_PTR(err);
2871         }
2872
2873         dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig);
2874         if (IS_ERR(dst)) {
2875                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
2876                 return ERR_CAST(dst);
2877         }
2878
2879         xdst = (struct xfrm_dst *)dst;
2880         xdst->num_xfrms = err;
2881         xdst->num_pols = num_pols;
2882         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
2883         xdst->policy_genid = atomic_read(&pols[0]->genid);
2884
2885         return xdst;
2886 }
2887
2888 static void xfrm_policy_queue_process(struct timer_list *t)
2889 {
2890         struct sk_buff *skb;
2891         struct sock *sk;
2892         struct dst_entry *dst;
2893         struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer);
2894         struct net *net = xp_net(pol);
2895         struct xfrm_policy_queue *pq = &pol->polq;
2896         struct flowi fl;
2897         struct sk_buff_head list;
2898         __u32 skb_mark;
2899
2900         spin_lock(&pq->hold_queue.lock);
2901         skb = skb_peek(&pq->hold_queue);
2902         if (!skb) {
2903                 spin_unlock(&pq->hold_queue.lock);
2904                 goto out;
2905         }
2906         dst = skb_dst(skb);
2907         sk = skb->sk;
2908
2909         /* Fixup the mark to support VTI. */
2910         skb_mark = skb->mark;
2911         skb->mark = pol->mark.v;
2912         xfrm_decode_session(net, skb, &fl, dst->ops->family);
2913         skb->mark = skb_mark;
2914         spin_unlock(&pq->hold_queue.lock);
2915
2916         dst_hold(xfrm_dst_path(dst));
2917         dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE);
2918         if (IS_ERR(dst))
2919                 goto purge_queue;
2920
2921         if (dst->flags & DST_XFRM_QUEUE) {
2922                 dst_release(dst);
2923
2924                 if (pq->timeout >= XFRM_QUEUE_TMO_MAX)
2925                         goto purge_queue;
2926
2927                 pq->timeout = pq->timeout << 1;
2928                 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout))
2929                         xfrm_pol_hold(pol);
2930                 goto out;
2931         }
2932
2933         dst_release(dst);
2934
2935         __skb_queue_head_init(&list);
2936
2937         spin_lock(&pq->hold_queue.lock);
2938         pq->timeout = 0;
2939         skb_queue_splice_init(&pq->hold_queue, &list);
2940         spin_unlock(&pq->hold_queue.lock);
2941
2942         while (!skb_queue_empty(&list)) {
2943                 skb = __skb_dequeue(&list);
2944
2945                 /* Fixup the mark to support VTI. */
2946                 skb_mark = skb->mark;
2947                 skb->mark = pol->mark.v;
2948                 xfrm_decode_session(net, skb, &fl, skb_dst(skb)->ops->family);
2949                 skb->mark = skb_mark;
2950
2951                 dst_hold(xfrm_dst_path(skb_dst(skb)));
2952                 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0);
2953                 if (IS_ERR(dst)) {
2954                         kfree_skb(skb);
2955                         continue;
2956                 }
2957
2958                 nf_reset_ct(skb);
2959                 skb_dst_drop(skb);
2960                 skb_dst_set(skb, dst);
2961
2962                 dst_output(net, skb->sk, skb);
2963         }
2964
2965 out:
2966         xfrm_pol_put(pol);
2967         return;
2968
2969 purge_queue:
2970         pq->timeout = 0;
2971         skb_queue_purge(&pq->hold_queue);
2972         xfrm_pol_put(pol);
2973 }
2974
2975 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb)
2976 {
2977         unsigned long sched_next;
2978         struct dst_entry *dst = skb_dst(skb);
2979         struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
2980         struct xfrm_policy *pol = xdst->pols[0];
2981         struct xfrm_policy_queue *pq = &pol->polq;
2982
2983         if (unlikely(skb_fclone_busy(sk, skb))) {
2984                 kfree_skb(skb);
2985                 return 0;
2986         }
2987
2988         if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) {
2989                 kfree_skb(skb);
2990                 return -EAGAIN;
2991         }
2992
2993         skb_dst_force(skb);
2994
2995         spin_lock_bh(&pq->hold_queue.lock);
2996
2997         if (!pq->timeout)
2998                 pq->timeout = XFRM_QUEUE_TMO_MIN;
2999
3000         sched_next = jiffies + pq->timeout;
3001
3002         if (del_timer(&pq->hold_timer)) {
3003                 if (time_before(pq->hold_timer.expires, sched_next))
3004                         sched_next = pq->hold_timer.expires;
3005                 xfrm_pol_put(pol);
3006         }
3007
3008         __skb_queue_tail(&pq->hold_queue, skb);
3009         if (!mod_timer(&pq->hold_timer, sched_next))
3010                 xfrm_pol_hold(pol);
3011
3012         spin_unlock_bh(&pq->hold_queue.lock);
3013
3014         return 0;
3015 }
3016
3017 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net,
3018                                                  struct xfrm_flo *xflo,
3019                                                  const struct flowi *fl,
3020                                                  int num_xfrms,
3021                                                  u16 family)
3022 {
3023         int err;
3024         struct net_device *dev;
3025         struct dst_entry *dst;
3026         struct dst_entry *dst1;
3027         struct xfrm_dst *xdst;
3028
3029         xdst = xfrm_alloc_dst(net, family);
3030         if (IS_ERR(xdst))
3031                 return xdst;
3032
3033         if (!(xflo->flags & XFRM_LOOKUP_QUEUE) ||
3034             net->xfrm.sysctl_larval_drop ||
3035             num_xfrms <= 0)
3036                 return xdst;
3037
3038         dst = xflo->dst_orig;
3039         dst1 = &xdst->u.dst;
3040         dst_hold(dst);
3041         xdst->route = dst;
3042
3043         dst_copy_metrics(dst1, dst);
3044
3045         dst1->obsolete = DST_OBSOLETE_FORCE_CHK;
3046         dst1->flags |= DST_XFRM_QUEUE;
3047         dst1->lastuse = jiffies;
3048
3049         dst1->input = dst_discard;
3050         dst1->output = xdst_queue_output;
3051
3052         dst_hold(dst);
3053         xfrm_dst_set_child(xdst, dst);
3054         xdst->path = dst;
3055
3056         xfrm_init_path((struct xfrm_dst *)dst1, dst, 0);
3057
3058         err = -ENODEV;
3059         dev = dst->dev;
3060         if (!dev)
3061                 goto free_dst;
3062
3063         err = xfrm_fill_dst(xdst, dev, fl);
3064         if (err)
3065                 goto free_dst;
3066
3067 out:
3068         return xdst;
3069
3070 free_dst:
3071         dst_release(dst1);
3072         xdst = ERR_PTR(err);
3073         goto out;
3074 }
3075
3076 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net,
3077                                            const struct flowi *fl,
3078                                            u16 family, u8 dir,
3079                                            struct xfrm_flo *xflo, u32 if_id)
3080 {
3081         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3082         int num_pols = 0, num_xfrms = 0, err;
3083         struct xfrm_dst *xdst;
3084
3085         /* Resolve policies to use if we couldn't get them from
3086          * previous cache entry */
3087         num_pols = 1;
3088         pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id);
3089         err = xfrm_expand_policies(fl, family, pols,
3090                                            &num_pols, &num_xfrms);
3091         if (err < 0)
3092                 goto inc_error;
3093         if (num_pols == 0)
3094                 return NULL;
3095         if (num_xfrms <= 0)
3096                 goto make_dummy_bundle;
3097
3098         xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family,
3099                                               xflo->dst_orig);
3100         if (IS_ERR(xdst)) {
3101                 err = PTR_ERR(xdst);
3102                 if (err == -EREMOTE) {
3103                         xfrm_pols_put(pols, num_pols);
3104                         return NULL;
3105                 }
3106
3107                 if (err != -EAGAIN)
3108                         goto error;
3109                 goto make_dummy_bundle;
3110         } else if (xdst == NULL) {
3111                 num_xfrms = 0;
3112                 goto make_dummy_bundle;
3113         }
3114
3115         return xdst;
3116
3117 make_dummy_bundle:
3118         /* We found policies, but there's no bundles to instantiate:
3119          * either because the policy blocks, has no transformations or
3120          * we could not build template (no xfrm_states).*/
3121         xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family);
3122         if (IS_ERR(xdst)) {
3123                 xfrm_pols_put(pols, num_pols);
3124                 return ERR_CAST(xdst);
3125         }
3126         xdst->num_pols = num_pols;
3127         xdst->num_xfrms = num_xfrms;
3128         memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols);
3129
3130         return xdst;
3131
3132 inc_error:
3133         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
3134 error:
3135         xfrm_pols_put(pols, num_pols);
3136         return ERR_PTR(err);
3137 }
3138
3139 static struct dst_entry *make_blackhole(struct net *net, u16 family,
3140                                         struct dst_entry *dst_orig)
3141 {
3142         const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
3143         struct dst_entry *ret;
3144
3145         if (!afinfo) {
3146                 dst_release(dst_orig);
3147                 return ERR_PTR(-EINVAL);
3148         } else {
3149                 ret = afinfo->blackhole_route(net, dst_orig);
3150         }
3151         rcu_read_unlock();
3152
3153         return ret;
3154 }
3155
3156 /* Finds/creates a bundle for given flow and if_id
3157  *
3158  * At the moment we eat a raw IP route. Mostly to speed up lookups
3159  * on interfaces with disabled IPsec.
3160  *
3161  * xfrm_lookup uses an if_id of 0 by default, and is provided for
3162  * compatibility
3163  */
3164 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
3165                                         struct dst_entry *dst_orig,
3166                                         const struct flowi *fl,
3167                                         const struct sock *sk,
3168                                         int flags, u32 if_id)
3169 {
3170         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3171         struct xfrm_dst *xdst;
3172         struct dst_entry *dst, *route;
3173         u16 family = dst_orig->ops->family;
3174         u8 dir = XFRM_POLICY_OUT;
3175         int i, err, num_pols, num_xfrms = 0, drop_pols = 0;
3176
3177         dst = NULL;
3178         xdst = NULL;
3179         route = NULL;
3180
3181         sk = sk_const_to_full_sk(sk);
3182         if (sk && sk->sk_policy[XFRM_POLICY_OUT]) {
3183                 num_pols = 1;
3184                 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family,
3185                                                 if_id);
3186                 err = xfrm_expand_policies(fl, family, pols,
3187                                            &num_pols, &num_xfrms);
3188                 if (err < 0)
3189                         goto dropdst;
3190
3191                 if (num_pols) {
3192                         if (num_xfrms <= 0) {
3193                                 drop_pols = num_pols;
3194                                 goto no_transform;
3195                         }
3196
3197                         xdst = xfrm_resolve_and_create_bundle(
3198                                         pols, num_pols, fl,
3199                                         family, dst_orig);
3200
3201                         if (IS_ERR(xdst)) {
3202                                 xfrm_pols_put(pols, num_pols);
3203                                 err = PTR_ERR(xdst);
3204                                 if (err == -EREMOTE)
3205                                         goto nopol;
3206
3207                                 goto dropdst;
3208                         } else if (xdst == NULL) {
3209                                 num_xfrms = 0;
3210                                 drop_pols = num_pols;
3211                                 goto no_transform;
3212                         }
3213
3214                         route = xdst->route;
3215                 }
3216         }
3217
3218         if (xdst == NULL) {
3219                 struct xfrm_flo xflo;
3220
3221                 xflo.dst_orig = dst_orig;
3222                 xflo.flags = flags;
3223
3224                 /* To accelerate a bit...  */
3225                 if (!if_id && ((dst_orig->flags & DST_NOXFRM) ||
3226                                !net->xfrm.policy_count[XFRM_POLICY_OUT]))
3227                         goto nopol;
3228
3229                 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id);
3230                 if (xdst == NULL)
3231                         goto nopol;
3232                 if (IS_ERR(xdst)) {
3233                         err = PTR_ERR(xdst);
3234                         goto dropdst;
3235                 }
3236
3237                 num_pols = xdst->num_pols;
3238                 num_xfrms = xdst->num_xfrms;
3239                 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols);
3240                 route = xdst->route;
3241         }
3242
3243         dst = &xdst->u.dst;
3244         if (route == NULL && num_xfrms > 0) {
3245                 /* The only case when xfrm_bundle_lookup() returns a
3246                  * bundle with null route, is when the template could
3247                  * not be resolved. It means policies are there, but
3248                  * bundle could not be created, since we don't yet
3249                  * have the xfrm_state's. We need to wait for KM to
3250                  * negotiate new SA's or bail out with error.*/
3251                 if (net->xfrm.sysctl_larval_drop) {
3252                         XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3253                         err = -EREMOTE;
3254                         goto error;
3255                 }
3256
3257                 err = -EAGAIN;
3258
3259                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
3260                 goto error;
3261         }
3262
3263 no_transform:
3264         if (num_pols == 0)
3265                 goto nopol;
3266
3267         if ((flags & XFRM_LOOKUP_ICMP) &&
3268             !(pols[0]->flags & XFRM_POLICY_ICMP)) {
3269                 err = -ENOENT;
3270                 goto error;
3271         }
3272
3273         for (i = 0; i < num_pols; i++)
3274                 WRITE_ONCE(pols[i]->curlft.use_time, ktime_get_real_seconds());
3275
3276         if (num_xfrms < 0) {
3277                 /* Prohibit the flow */
3278                 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
3279                 err = -EPERM;
3280                 goto error;
3281         } else if (num_xfrms > 0) {
3282                 /* Flow transformed */
3283                 dst_release(dst_orig);
3284         } else {
3285                 /* Flow passes untransformed */
3286                 dst_release(dst);
3287                 dst = dst_orig;
3288         }
3289
3290 ok:
3291         xfrm_pols_put(pols, drop_pols);
3292         if (dst && dst->xfrm &&
3293             dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
3294                 dst->flags |= DST_XFRM_TUNNEL;
3295         return dst;
3296
3297 nopol:
3298         if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) &&
3299             net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3300                 err = -EPERM;
3301                 goto error;
3302         }
3303         if (!(flags & XFRM_LOOKUP_ICMP)) {
3304                 dst = dst_orig;
3305                 goto ok;
3306         }
3307         err = -ENOENT;
3308 error:
3309         dst_release(dst);
3310 dropdst:
3311         if (!(flags & XFRM_LOOKUP_KEEP_DST_REF))
3312                 dst_release(dst_orig);
3313         xfrm_pols_put(pols, drop_pols);
3314         return ERR_PTR(err);
3315 }
3316 EXPORT_SYMBOL(xfrm_lookup_with_ifid);
3317
3318 /* Main function: finds/creates a bundle for given flow.
3319  *
3320  * At the moment we eat a raw IP route. Mostly to speed up lookups
3321  * on interfaces with disabled IPsec.
3322  */
3323 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
3324                               const struct flowi *fl, const struct sock *sk,
3325                               int flags)
3326 {
3327         return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0);
3328 }
3329 EXPORT_SYMBOL(xfrm_lookup);
3330
3331 /* Callers of xfrm_lookup_route() must ensure a call to dst_output().
3332  * Otherwise we may send out blackholed packets.
3333  */
3334 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
3335                                     const struct flowi *fl,
3336                                     const struct sock *sk, int flags)
3337 {
3338         struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk,
3339                                             flags | XFRM_LOOKUP_QUEUE |
3340                                             XFRM_LOOKUP_KEEP_DST_REF);
3341
3342         if (PTR_ERR(dst) == -EREMOTE)
3343                 return make_blackhole(net, dst_orig->ops->family, dst_orig);
3344
3345         if (IS_ERR(dst))
3346                 dst_release(dst_orig);
3347
3348         return dst;
3349 }
3350 EXPORT_SYMBOL(xfrm_lookup_route);
3351
3352 static inline int
3353 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl)
3354 {
3355         struct sec_path *sp = skb_sec_path(skb);
3356         struct xfrm_state *x;
3357
3358         if (!sp || idx < 0 || idx >= sp->len)
3359                 return 0;
3360         x = sp->xvec[idx];
3361         if (!x->type->reject)
3362                 return 0;
3363         return x->type->reject(x, skb, fl);
3364 }
3365
3366 /* When skb is transformed back to its "native" form, we have to
3367  * check policy restrictions. At the moment we make this in maximally
3368  * stupid way. Shame on me. :-) Of course, connected sockets must
3369  * have policy cached at them.
3370  */
3371
3372 static inline int
3373 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x,
3374               unsigned short family, u32 if_id)
3375 {
3376         if (xfrm_state_kern(x))
3377                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family);
3378         return  x->id.proto == tmpl->id.proto &&
3379                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
3380                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
3381                 x->props.mode == tmpl->mode &&
3382                 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) ||
3383                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
3384                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
3385                   xfrm_state_addr_cmp(tmpl, x, family)) &&
3386                 (if_id == 0 || if_id == x->if_id);
3387 }
3388
3389 /*
3390  * 0 or more than 0 is returned when validation is succeeded (either bypass
3391  * because of optional transport mode, or next index of the matched secpath
3392  * state with the template.
3393  * -1 is returned when no matching template is found.
3394  * Otherwise "-2 - errored_index" is returned.
3395  */
3396 static inline int
3397 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start,
3398                unsigned short family, u32 if_id)
3399 {
3400         int idx = start;
3401
3402         if (tmpl->optional) {
3403                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
3404                         return start;
3405         } else
3406                 start = -1;
3407         for (; idx < sp->len; idx++) {
3408                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id))
3409                         return ++idx;
3410                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
3411                         if (idx < sp->verified_cnt) {
3412                                 /* Secpath entry previously verified, consider optional and
3413                                  * continue searching
3414                                  */
3415                                 continue;
3416                         }
3417
3418                         if (start == -1)
3419                                 start = -2-idx;
3420                         break;
3421                 }
3422         }
3423         return start;
3424 }
3425
3426 static void
3427 decode_session4(const struct xfrm_flow_keys *flkeys, struct flowi *fl, bool reverse)
3428 {
3429         struct flowi4 *fl4 = &fl->u.ip4;
3430
3431         memset(fl4, 0, sizeof(struct flowi4));
3432
3433         if (reverse) {
3434                 fl4->saddr = flkeys->addrs.ipv4.dst;
3435                 fl4->daddr = flkeys->addrs.ipv4.src;
3436                 fl4->fl4_sport = flkeys->ports.dst;
3437                 fl4->fl4_dport = flkeys->ports.src;
3438         } else {
3439                 fl4->saddr = flkeys->addrs.ipv4.src;
3440                 fl4->daddr = flkeys->addrs.ipv4.dst;
3441                 fl4->fl4_sport = flkeys->ports.src;
3442                 fl4->fl4_dport = flkeys->ports.dst;
3443         }
3444
3445         switch (flkeys->basic.ip_proto) {
3446         case IPPROTO_GRE:
3447                 fl4->fl4_gre_key = flkeys->gre.keyid;
3448                 break;
3449         case IPPROTO_ICMP:
3450                 fl4->fl4_icmp_type = flkeys->icmp.type;
3451                 fl4->fl4_icmp_code = flkeys->icmp.code;
3452                 break;
3453         }
3454
3455         fl4->flowi4_proto = flkeys->basic.ip_proto;
3456         fl4->flowi4_tos = flkeys->ip.tos & ~INET_ECN_MASK;
3457 }
3458
3459 #if IS_ENABLED(CONFIG_IPV6)
3460 static void
3461 decode_session6(const struct xfrm_flow_keys *flkeys, struct flowi *fl, bool reverse)
3462 {
3463         struct flowi6 *fl6 = &fl->u.ip6;
3464
3465         memset(fl6, 0, sizeof(struct flowi6));
3466
3467         if (reverse) {
3468                 fl6->saddr = flkeys->addrs.ipv6.dst;
3469                 fl6->daddr = flkeys->addrs.ipv6.src;
3470                 fl6->fl6_sport = flkeys->ports.dst;
3471                 fl6->fl6_dport = flkeys->ports.src;
3472         } else {
3473                 fl6->saddr = flkeys->addrs.ipv6.src;
3474                 fl6->daddr = flkeys->addrs.ipv6.dst;
3475                 fl6->fl6_sport = flkeys->ports.src;
3476                 fl6->fl6_dport = flkeys->ports.dst;
3477         }
3478
3479         switch (flkeys->basic.ip_proto) {
3480         case IPPROTO_GRE:
3481                 fl6->fl6_gre_key = flkeys->gre.keyid;
3482                 break;
3483         case IPPROTO_ICMPV6:
3484                 fl6->fl6_icmp_type = flkeys->icmp.type;
3485                 fl6->fl6_icmp_code = flkeys->icmp.code;
3486                 break;
3487         }
3488
3489         fl6->flowi6_proto = flkeys->basic.ip_proto;
3490 }
3491 #endif
3492
3493 int __xfrm_decode_session(struct net *net, struct sk_buff *skb, struct flowi *fl,
3494                           unsigned int family, int reverse)
3495 {
3496         struct xfrm_flow_keys flkeys;
3497
3498         memset(&flkeys, 0, sizeof(flkeys));
3499         __skb_flow_dissect(net, skb, &xfrm_session_dissector, &flkeys,
3500                            NULL, 0, 0, 0, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
3501
3502         switch (family) {
3503         case AF_INET:
3504                 decode_session4(&flkeys, fl, reverse);
3505                 break;
3506 #if IS_ENABLED(CONFIG_IPV6)
3507         case AF_INET6:
3508                 decode_session6(&flkeys, fl, reverse);
3509                 break;
3510 #endif
3511         default:
3512                 return -EAFNOSUPPORT;
3513         }
3514
3515         fl->flowi_mark = skb->mark;
3516         if (reverse) {
3517                 fl->flowi_oif = skb->skb_iif;
3518         } else {
3519                 int oif = 0;
3520
3521                 if (skb_dst(skb) && skb_dst(skb)->dev)
3522                         oif = skb_dst(skb)->dev->ifindex;
3523
3524                 fl->flowi_oif = oif;
3525         }
3526
3527         return security_xfrm_decode_session(skb, &fl->flowi_secid);
3528 }
3529 EXPORT_SYMBOL(__xfrm_decode_session);
3530
3531 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp)
3532 {
3533         for (; k < sp->len; k++) {
3534                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
3535                         *idxp = k;
3536                         return 1;
3537                 }
3538         }
3539
3540         return 0;
3541 }
3542
3543 static bool icmp_err_packet(const struct flowi *fl, unsigned short family)
3544 {
3545         const struct flowi4 *fl4 = &fl->u.ip4;
3546
3547         if (family == AF_INET &&
3548             fl4->flowi4_proto == IPPROTO_ICMP &&
3549             (fl4->fl4_icmp_type == ICMP_DEST_UNREACH ||
3550              fl4->fl4_icmp_type == ICMP_TIME_EXCEEDED))
3551                 return true;
3552
3553 #if IS_ENABLED(CONFIG_IPV6)
3554         if (family == AF_INET6) {
3555                 const struct flowi6 *fl6 = &fl->u.ip6;
3556
3557                 if (fl6->flowi6_proto == IPPROTO_ICMPV6 &&
3558                     (fl6->fl6_icmp_type == ICMPV6_DEST_UNREACH ||
3559                     fl6->fl6_icmp_type == ICMPV6_PKT_TOOBIG ||
3560                     fl6->fl6_icmp_type == ICMPV6_TIME_EXCEED))
3561                         return true;
3562         }
3563 #endif
3564         return false;
3565 }
3566
3567 static bool xfrm_icmp_flow_decode(struct sk_buff *skb, unsigned short family,
3568                                   const struct flowi *fl, struct flowi *fl1)
3569 {
3570         bool ret = true;
3571         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
3572         int hl = family == AF_INET ? (sizeof(struct iphdr) +  sizeof(struct icmphdr)) :
3573                  (sizeof(struct ipv6hdr) + sizeof(struct icmp6hdr));
3574
3575         if (!newskb)
3576                 return true;
3577
3578         if (!pskb_pull(newskb, hl))
3579                 goto out;
3580
3581         skb_reset_network_header(newskb);
3582
3583         if (xfrm_decode_session_reverse(dev_net(skb->dev), newskb, fl1, family) < 0)
3584                 goto out;
3585
3586         fl1->flowi_oif = fl->flowi_oif;
3587         fl1->flowi_mark = fl->flowi_mark;
3588         fl1->flowi_tos = fl->flowi_tos;
3589         nf_nat_decode_session(newskb, fl1, family);
3590         ret = false;
3591
3592 out:
3593         consume_skb(newskb);
3594         return ret;
3595 }
3596
3597 static bool xfrm_selector_inner_icmp_match(struct sk_buff *skb, unsigned short family,
3598                                            const struct xfrm_selector *sel,
3599                                            const struct flowi *fl)
3600 {
3601         bool ret = false;
3602
3603         if (icmp_err_packet(fl, family)) {
3604                 struct flowi fl1;
3605
3606                 if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
3607                         return ret;
3608
3609                 ret = xfrm_selector_match(sel, &fl1, family);
3610         }
3611
3612         return ret;
3613 }
3614
3615 static inline struct
3616 xfrm_policy *xfrm_in_fwd_icmp(struct sk_buff *skb,
3617                               const struct flowi *fl, unsigned short family,
3618                               u32 if_id)
3619 {
3620         struct xfrm_policy *pol = NULL;
3621
3622         if (icmp_err_packet(fl, family)) {
3623                 struct flowi fl1;
3624                 struct net *net = dev_net(skb->dev);
3625
3626                 if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
3627                         return pol;
3628
3629                 pol = xfrm_policy_lookup(net, &fl1, family, XFRM_POLICY_FWD, if_id);
3630                 if (IS_ERR(pol))
3631                         pol = NULL;
3632         }
3633
3634         return pol;
3635 }
3636
3637 static inline struct
3638 dst_entry *xfrm_out_fwd_icmp(struct sk_buff *skb, struct flowi *fl,
3639                              unsigned short family, struct dst_entry *dst)
3640 {
3641         if (icmp_err_packet(fl, family)) {
3642                 struct net *net = dev_net(skb->dev);
3643                 struct dst_entry *dst2;
3644                 struct flowi fl1;
3645
3646                 if (xfrm_icmp_flow_decode(skb, family, fl, &fl1))
3647                         return dst;
3648
3649                 dst_hold(dst);
3650
3651                 dst2 = xfrm_lookup(net, dst, &fl1, NULL, (XFRM_LOOKUP_QUEUE | XFRM_LOOKUP_ICMP));
3652
3653                 if (IS_ERR(dst2))
3654                         return dst;
3655
3656                 if (dst2->xfrm) {
3657                         dst_release(dst);
3658                         dst = dst2;
3659                 } else {
3660                         dst_release(dst2);
3661                 }
3662         }
3663
3664         return dst;
3665 }
3666
3667 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
3668                         unsigned short family)
3669 {
3670         struct net *net = dev_net(skb->dev);
3671         struct xfrm_policy *pol;
3672         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
3673         int npols = 0;
3674         int xfrm_nr;
3675         int pi;
3676         int reverse;
3677         struct flowi fl;
3678         int xerr_idx = -1;
3679         const struct xfrm_if_cb *ifcb;
3680         struct sec_path *sp;
3681         u32 if_id = 0;
3682
3683         rcu_read_lock();
3684         ifcb = xfrm_if_get_cb();
3685
3686         if (ifcb) {
3687                 struct xfrm_if_decode_session_result r;
3688
3689                 if (ifcb->decode_session(skb, family, &r)) {
3690                         if_id = r.if_id;
3691                         net = r.net;
3692                 }
3693         }
3694         rcu_read_unlock();
3695
3696         reverse = dir & ~XFRM_POLICY_MASK;
3697         dir &= XFRM_POLICY_MASK;
3698
3699         if (__xfrm_decode_session(net, skb, &fl, family, reverse) < 0) {
3700                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
3701                 return 0;
3702         }
3703
3704         nf_nat_decode_session(skb, &fl, family);
3705
3706         /* First, check used SA against their selectors. */
3707         sp = skb_sec_path(skb);
3708         if (sp) {
3709                 int i;
3710
3711                 for (i = sp->len - 1; i >= 0; i--) {
3712                         struct xfrm_state *x = sp->xvec[i];
3713                         int ret = 0;
3714
3715                         if (!xfrm_selector_match(&x->sel, &fl, family)) {
3716                                 ret = 1;
3717                                 if (x->props.flags & XFRM_STATE_ICMP &&
3718                                     xfrm_selector_inner_icmp_match(skb, family, &x->sel, &fl))
3719                                         ret = 0;
3720                                 if (ret) {
3721                                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
3722                                         return 0;
3723                                 }
3724                         }
3725                 }
3726         }
3727
3728         pol = NULL;
3729         sk = sk_to_full_sk(sk);
3730         if (sk && sk->sk_policy[dir]) {
3731                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id);
3732                 if (IS_ERR(pol)) {
3733                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3734                         return 0;
3735                 }
3736         }
3737
3738         if (!pol)
3739                 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id);
3740
3741         if (IS_ERR(pol)) {
3742                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3743                 return 0;
3744         }
3745
3746         if (!pol && dir == XFRM_POLICY_FWD)
3747                 pol = xfrm_in_fwd_icmp(skb, &fl, family, if_id);
3748
3749         if (!pol) {
3750                 const bool is_crypto_offload = sp &&
3751                         (xfrm_input_state(skb)->xso.type == XFRM_DEV_OFFLOAD_CRYPTO);
3752
3753                 if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) {
3754                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3755                         return 0;
3756                 }
3757
3758                 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx) && !is_crypto_offload) {
3759                         xfrm_secpath_reject(xerr_idx, skb, &fl);
3760                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
3761                         return 0;
3762                 }
3763                 return 1;
3764         }
3765
3766         /* This lockless write can happen from different cpus. */
3767         WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds());
3768
3769         pols[0] = pol;
3770         npols++;
3771 #ifdef CONFIG_XFRM_SUB_POLICY
3772         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
3773                 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
3774                                                     &fl, family,
3775                                                     XFRM_POLICY_IN, if_id);
3776                 if (pols[1]) {
3777                         if (IS_ERR(pols[1])) {
3778                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
3779                                 xfrm_pol_put(pols[0]);
3780                                 return 0;
3781                         }
3782                         /* This write can happen from different cpus. */
3783                         WRITE_ONCE(pols[1]->curlft.use_time,
3784                                    ktime_get_real_seconds());
3785                         npols++;
3786                 }
3787         }
3788 #endif
3789
3790         if (pol->action == XFRM_POLICY_ALLOW) {
3791                 static struct sec_path dummy;
3792                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
3793                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
3794                 struct xfrm_tmpl **tpp = tp;
3795                 int ti = 0;
3796                 int i, k;
3797
3798                 sp = skb_sec_path(skb);
3799                 if (!sp)
3800                         sp = &dummy;
3801
3802                 for (pi = 0; pi < npols; pi++) {
3803                         if (pols[pi] != pol &&
3804                             pols[pi]->action != XFRM_POLICY_ALLOW) {
3805                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3806                                 goto reject;
3807                         }
3808                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
3809                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
3810                                 goto reject_error;
3811                         }
3812                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
3813                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
3814                 }
3815                 xfrm_nr = ti;
3816
3817                 if (npols > 1) {
3818                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
3819                         tpp = stp;
3820                 }
3821
3822                 /* For each tunnel xfrm, find the first matching tmpl.
3823                  * For each tmpl before that, find corresponding xfrm.
3824                  * Order is _important_. Later we will implement
3825                  * some barriers, but at the moment barriers
3826                  * are implied between each two transformations.
3827                  * Upon success, marks secpath entries as having been
3828                  * verified to allow them to be skipped in future policy
3829                  * checks (e.g. nested tunnels).
3830                  */
3831                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
3832                         k = xfrm_policy_ok(tpp[i], sp, k, family, if_id);
3833                         if (k < 0) {
3834                                 if (k < -1)
3835                                         /* "-2 - errored_index" returned */
3836                                         xerr_idx = -(2+k);
3837                                 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3838                                 goto reject;
3839                         }
3840                 }
3841
3842                 if (secpath_has_nontransport(sp, k, &xerr_idx)) {
3843                         XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
3844                         goto reject;
3845                 }
3846
3847                 xfrm_pols_put(pols, npols);
3848                 sp->verified_cnt = k;
3849
3850                 return 1;
3851         }
3852         XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
3853
3854 reject:
3855         xfrm_secpath_reject(xerr_idx, skb, &fl);
3856 reject_error:
3857         xfrm_pols_put(pols, npols);
3858         return 0;
3859 }
3860 EXPORT_SYMBOL(__xfrm_policy_check);
3861
3862 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
3863 {
3864         struct net *net = dev_net(skb->dev);
3865         struct flowi fl;
3866         struct dst_entry *dst;
3867         int res = 1;
3868
3869         if (xfrm_decode_session(net, skb, &fl, family) < 0) {
3870                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3871                 return 0;
3872         }
3873
3874         skb_dst_force(skb);
3875         if (!skb_dst(skb)) {
3876                 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR);
3877                 return 0;
3878         }
3879
3880         dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE);
3881         if (IS_ERR(dst)) {
3882                 res = 0;
3883                 dst = NULL;
3884         }
3885
3886         if (dst && !dst->xfrm)
3887                 dst = xfrm_out_fwd_icmp(skb, &fl, family, dst);
3888
3889         skb_dst_set(skb, dst);
3890         return res;
3891 }
3892 EXPORT_SYMBOL(__xfrm_route_forward);
3893
3894 /* Optimize later using cookies and generation ids. */
3895
3896 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
3897 {
3898         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
3899          * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to
3900          * get validated by dst_ops->check on every use.  We do this
3901          * because when a normal route referenced by an XFRM dst is
3902          * obsoleted we do not go looking around for all parent
3903          * referencing XFRM dsts so that we can invalidate them.  It
3904          * is just too much work.  Instead we make the checks here on
3905          * every use.  For example:
3906          *
3907          *      XFRM dst A --> IPv4 dst X
3908          *
3909          * X is the "xdst->route" of A (X is also the "dst->path" of A
3910          * in this example).  If X is marked obsolete, "A" will not
3911          * notice.  That's what we are validating here via the
3912          * stale_bundle() check.
3913          *
3914          * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will
3915          * be marked on it.
3916          * This will force stale_bundle() to fail on any xdst bundle with
3917          * this dst linked in it.
3918          */
3919         if (dst->obsolete < 0 && !stale_bundle(dst))
3920                 return dst;
3921
3922         return NULL;
3923 }
3924
3925 static int stale_bundle(struct dst_entry *dst)
3926 {
3927         return !xfrm_bundle_ok((struct xfrm_dst *)dst);
3928 }
3929
3930 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
3931 {
3932         while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) {
3933                 dst->dev = blackhole_netdev;
3934                 dev_hold(dst->dev);
3935                 dev_put(dev);
3936         }
3937 }
3938 EXPORT_SYMBOL(xfrm_dst_ifdown);
3939
3940 static void xfrm_link_failure(struct sk_buff *skb)
3941 {
3942         /* Impossible. Such dst must be popped before reaches point of failure. */
3943 }
3944
3945 static void xfrm_negative_advice(struct sock *sk, struct dst_entry *dst)
3946 {
3947         if (dst->obsolete)
3948                 sk_dst_reset(sk);
3949 }
3950
3951 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr)
3952 {
3953         while (nr--) {
3954                 struct xfrm_dst *xdst = bundle[nr];
3955                 u32 pmtu, route_mtu_cached;
3956                 struct dst_entry *dst;
3957
3958                 dst = &xdst->u.dst;
3959                 pmtu = dst_mtu(xfrm_dst_child(dst));
3960                 xdst->child_mtu_cached = pmtu;
3961
3962                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
3963
3964                 route_mtu_cached = dst_mtu(xdst->route);
3965                 xdst->route_mtu_cached = route_mtu_cached;
3966
3967                 if (pmtu > route_mtu_cached)
3968                         pmtu = route_mtu_cached;
3969
3970                 dst_metric_set(dst, RTAX_MTU, pmtu);
3971         }
3972 }
3973
3974 /* Check that the bundle accepts the flow and its components are
3975  * still valid.
3976  */
3977
3978 static int xfrm_bundle_ok(struct xfrm_dst *first)
3979 {
3980         struct xfrm_dst *bundle[XFRM_MAX_DEPTH];
3981         struct dst_entry *dst = &first->u.dst;
3982         struct xfrm_dst *xdst;
3983         int start_from, nr;
3984         u32 mtu;
3985
3986         if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) ||
3987             (dst->dev && !netif_running(dst->dev)))
3988                 return 0;
3989
3990         if (dst->flags & DST_XFRM_QUEUE)
3991                 return 1;
3992
3993         start_from = nr = 0;
3994         do {
3995                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
3996
3997                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
3998                         return 0;
3999                 if (xdst->xfrm_genid != dst->xfrm->genid)
4000                         return 0;
4001                 if (xdst->num_pols > 0 &&
4002                     xdst->policy_genid != atomic_read(&xdst->pols[0]->genid))
4003                         return 0;
4004
4005                 bundle[nr++] = xdst;
4006
4007                 mtu = dst_mtu(xfrm_dst_child(dst));
4008                 if (xdst->child_mtu_cached != mtu) {
4009                         start_from = nr;
4010                         xdst->child_mtu_cached = mtu;
4011                 }
4012
4013                 if (!dst_check(xdst->route, xdst->route_cookie))
4014                         return 0;
4015                 mtu = dst_mtu(xdst->route);
4016                 if (xdst->route_mtu_cached != mtu) {
4017                         start_from = nr;
4018                         xdst->route_mtu_cached = mtu;
4019                 }
4020
4021                 dst = xfrm_dst_child(dst);
4022         } while (dst->xfrm);
4023
4024         if (likely(!start_from))
4025                 return 1;
4026
4027         xdst = bundle[start_from - 1];
4028         mtu = xdst->child_mtu_cached;
4029         while (start_from--) {
4030                 dst = &xdst->u.dst;
4031
4032                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
4033                 if (mtu > xdst->route_mtu_cached)
4034                         mtu = xdst->route_mtu_cached;
4035                 dst_metric_set(dst, RTAX_MTU, mtu);
4036                 if (!start_from)
4037                         break;
4038
4039                 xdst = bundle[start_from - 1];
4040                 xdst->child_mtu_cached = mtu;
4041         }
4042
4043         return 1;
4044 }
4045
4046 static unsigned int xfrm_default_advmss(const struct dst_entry *dst)
4047 {
4048         return dst_metric_advmss(xfrm_dst_path(dst));
4049 }
4050
4051 static unsigned int xfrm_mtu(const struct dst_entry *dst)
4052 {
4053         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
4054
4055         return mtu ? : dst_mtu(xfrm_dst_path(dst));
4056 }
4057
4058 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst,
4059                                         const void *daddr)
4060 {
4061         while (dst->xfrm) {
4062                 const struct xfrm_state *xfrm = dst->xfrm;
4063
4064                 dst = xfrm_dst_child(dst);
4065
4066                 if (xfrm->props.mode == XFRM_MODE_TRANSPORT)
4067                         continue;
4068                 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR)
4069                         daddr = xfrm->coaddr;
4070                 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR))
4071                         daddr = &xfrm->id.daddr;
4072         }
4073         return daddr;
4074 }
4075
4076 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst,
4077                                            struct sk_buff *skb,
4078                                            const void *daddr)
4079 {
4080         const struct dst_entry *path = xfrm_dst_path(dst);
4081
4082         if (!skb)
4083                 daddr = xfrm_get_dst_nexthop(dst, daddr);
4084         return path->ops->neigh_lookup(path, skb, daddr);
4085 }
4086
4087 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr)
4088 {
4089         const struct dst_entry *path = xfrm_dst_path(dst);
4090
4091         daddr = xfrm_get_dst_nexthop(dst, daddr);
4092         path->ops->confirm_neigh(path, daddr);
4093 }
4094
4095 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family)
4096 {
4097         int err = 0;
4098
4099         if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo)))
4100                 return -EAFNOSUPPORT;
4101
4102         spin_lock(&xfrm_policy_afinfo_lock);
4103         if (unlikely(xfrm_policy_afinfo[family] != NULL))
4104                 err = -EEXIST;
4105         else {
4106                 struct dst_ops *dst_ops = afinfo->dst_ops;
4107                 if (likely(dst_ops->kmem_cachep == NULL))
4108                         dst_ops->kmem_cachep = xfrm_dst_cache;
4109                 if (likely(dst_ops->check == NULL))
4110                         dst_ops->check = xfrm_dst_check;
4111                 if (likely(dst_ops->default_advmss == NULL))
4112                         dst_ops->default_advmss = xfrm_default_advmss;
4113                 if (likely(dst_ops->mtu == NULL))
4114                         dst_ops->mtu = xfrm_mtu;
4115                 if (likely(dst_ops->negative_advice == NULL))
4116                         dst_ops->negative_advice = xfrm_negative_advice;
4117                 if (likely(dst_ops->link_failure == NULL))
4118                         dst_ops->link_failure = xfrm_link_failure;
4119                 if (likely(dst_ops->neigh_lookup == NULL))
4120                         dst_ops->neigh_lookup = xfrm_neigh_lookup;
4121                 if (likely(!dst_ops->confirm_neigh))
4122                         dst_ops->confirm_neigh = xfrm_confirm_neigh;
4123                 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo);
4124         }
4125         spin_unlock(&xfrm_policy_afinfo_lock);
4126
4127         return err;
4128 }
4129 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
4130
4131 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo)
4132 {
4133         struct dst_ops *dst_ops = afinfo->dst_ops;
4134         int i;
4135
4136         for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) {
4137                 if (xfrm_policy_afinfo[i] != afinfo)
4138                         continue;
4139                 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL);
4140                 break;
4141         }
4142
4143         synchronize_rcu();
4144
4145         dst_ops->kmem_cachep = NULL;
4146         dst_ops->check = NULL;
4147         dst_ops->negative_advice = NULL;
4148         dst_ops->link_failure = NULL;
4149 }
4150 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
4151
4152 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb)
4153 {
4154         spin_lock(&xfrm_if_cb_lock);
4155         rcu_assign_pointer(xfrm_if_cb, ifcb);
4156         spin_unlock(&xfrm_if_cb_lock);
4157 }
4158 EXPORT_SYMBOL(xfrm_if_register_cb);
4159
4160 void xfrm_if_unregister_cb(void)
4161 {
4162         RCU_INIT_POINTER(xfrm_if_cb, NULL);
4163         synchronize_rcu();
4164 }
4165 EXPORT_SYMBOL(xfrm_if_unregister_cb);
4166
4167 #ifdef CONFIG_XFRM_STATISTICS
4168 static int __net_init xfrm_statistics_init(struct net *net)
4169 {
4170         int rv;
4171         net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib);
4172         if (!net->mib.xfrm_statistics)
4173                 return -ENOMEM;
4174         rv = xfrm_proc_init(net);
4175         if (rv < 0)
4176                 free_percpu(net->mib.xfrm_statistics);
4177         return rv;
4178 }
4179
4180 static void xfrm_statistics_fini(struct net *net)
4181 {
4182         xfrm_proc_fini(net);
4183         free_percpu(net->mib.xfrm_statistics);
4184 }
4185 #else
4186 static int __net_init xfrm_statistics_init(struct net *net)
4187 {
4188         return 0;
4189 }
4190
4191 static void xfrm_statistics_fini(struct net *net)
4192 {
4193 }
4194 #endif
4195
4196 static int __net_init xfrm_policy_init(struct net *net)
4197 {
4198         unsigned int hmask, sz;
4199         int dir, err;
4200
4201         if (net_eq(net, &init_net)) {
4202                 xfrm_dst_cache = KMEM_CACHE(xfrm_dst, SLAB_HWCACHE_ALIGN | SLAB_PANIC);
4203                 err = rhashtable_init(&xfrm_policy_inexact_table,
4204                                       &xfrm_pol_inexact_params);
4205                 BUG_ON(err);
4206         }
4207
4208         hmask = 8 - 1;
4209         sz = (hmask+1) * sizeof(struct hlist_head);
4210
4211         net->xfrm.policy_byidx = xfrm_hash_alloc(sz);
4212         if (!net->xfrm.policy_byidx)
4213                 goto out_byidx;
4214         net->xfrm.policy_idx_hmask = hmask;
4215
4216         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4217                 struct xfrm_policy_hash *htab;
4218
4219                 net->xfrm.policy_count[dir] = 0;
4220                 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0;
4221
4222                 htab = &net->xfrm.policy_bydst[dir];
4223                 htab->table = xfrm_hash_alloc(sz);
4224                 if (!htab->table)
4225                         goto out_bydst;
4226                 htab->hmask = hmask;
4227                 htab->dbits4 = 32;
4228                 htab->sbits4 = 32;
4229                 htab->dbits6 = 128;
4230                 htab->sbits6 = 128;
4231         }
4232         net->xfrm.policy_hthresh.lbits4 = 32;
4233         net->xfrm.policy_hthresh.rbits4 = 32;
4234         net->xfrm.policy_hthresh.lbits6 = 128;
4235         net->xfrm.policy_hthresh.rbits6 = 128;
4236
4237         seqlock_init(&net->xfrm.policy_hthresh.lock);
4238
4239         INIT_LIST_HEAD(&net->xfrm.policy_all);
4240         INIT_LIST_HEAD(&net->xfrm.inexact_bins);
4241         INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize);
4242         INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild);
4243         return 0;
4244
4245 out_bydst:
4246         for (dir--; dir >= 0; dir--) {
4247                 struct xfrm_policy_hash *htab;
4248
4249                 htab = &net->xfrm.policy_bydst[dir];
4250                 xfrm_hash_free(htab->table, sz);
4251         }
4252         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4253 out_byidx:
4254         return -ENOMEM;
4255 }
4256
4257 static void xfrm_policy_fini(struct net *net)
4258 {
4259         struct xfrm_pol_inexact_bin *b, *t;
4260         unsigned int sz;
4261         int dir;
4262
4263         flush_work(&net->xfrm.policy_hash_work);
4264 #ifdef CONFIG_XFRM_SUB_POLICY
4265         xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false);
4266 #endif
4267         xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false);
4268
4269         WARN_ON(!list_empty(&net->xfrm.policy_all));
4270
4271         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
4272                 struct xfrm_policy_hash *htab;
4273
4274                 htab = &net->xfrm.policy_bydst[dir];
4275                 sz = (htab->hmask + 1) * sizeof(struct hlist_head);
4276                 WARN_ON(!hlist_empty(htab->table));
4277                 xfrm_hash_free(htab->table, sz);
4278         }
4279
4280         sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head);
4281         WARN_ON(!hlist_empty(net->xfrm.policy_byidx));
4282         xfrm_hash_free(net->xfrm.policy_byidx, sz);
4283
4284         spin_lock_bh(&net->xfrm.xfrm_policy_lock);
4285         list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins)
4286                 __xfrm_policy_inexact_prune_bin(b, true);
4287         spin_unlock_bh(&net->xfrm.xfrm_policy_lock);
4288 }
4289
4290 static int __net_init xfrm_net_init(struct net *net)
4291 {
4292         int rv;
4293
4294         /* Initialize the per-net locks here */
4295         spin_lock_init(&net->xfrm.xfrm_state_lock);
4296         spin_lock_init(&net->xfrm.xfrm_policy_lock);
4297         seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock);
4298         mutex_init(&net->xfrm.xfrm_cfg_mutex);
4299         net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT;
4300         net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT;
4301         net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT;
4302
4303         rv = xfrm_statistics_init(net);
4304         if (rv < 0)
4305                 goto out_statistics;
4306         rv = xfrm_state_init(net);
4307         if (rv < 0)
4308                 goto out_state;
4309         rv = xfrm_policy_init(net);
4310         if (rv < 0)
4311                 goto out_policy;
4312         rv = xfrm_sysctl_init(net);
4313         if (rv < 0)
4314                 goto out_sysctl;
4315
4316         rv = xfrm_nat_keepalive_net_init(net);
4317         if (rv < 0)
4318                 goto out_nat_keepalive;
4319
4320         return 0;
4321
4322 out_nat_keepalive:
4323         xfrm_sysctl_fini(net);
4324 out_sysctl:
4325         xfrm_policy_fini(net);
4326 out_policy:
4327         xfrm_state_fini(net);
4328 out_state:
4329         xfrm_statistics_fini(net);
4330 out_statistics:
4331         return rv;
4332 }
4333
4334 static void __net_exit xfrm_net_exit(struct net *net)
4335 {
4336         xfrm_nat_keepalive_net_fini(net);
4337         xfrm_sysctl_fini(net);
4338         xfrm_policy_fini(net);
4339         xfrm_state_fini(net);
4340         xfrm_statistics_fini(net);
4341 }
4342
4343 static struct pernet_operations __net_initdata xfrm_net_ops = {
4344         .init = xfrm_net_init,
4345         .exit = xfrm_net_exit,
4346 };
4347
4348 static const struct flow_dissector_key xfrm_flow_dissector_keys[] = {
4349         {
4350                 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
4351                 .offset = offsetof(struct xfrm_flow_keys, control),
4352         },
4353         {
4354                 .key_id = FLOW_DISSECTOR_KEY_BASIC,
4355                 .offset = offsetof(struct xfrm_flow_keys, basic),
4356         },
4357         {
4358                 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
4359                 .offset = offsetof(struct xfrm_flow_keys, addrs.ipv4),
4360         },
4361         {
4362                 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
4363                 .offset = offsetof(struct xfrm_flow_keys, addrs.ipv6),
4364         },
4365         {
4366                 .key_id = FLOW_DISSECTOR_KEY_PORTS,
4367                 .offset = offsetof(struct xfrm_flow_keys, ports),
4368         },
4369         {
4370                 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
4371                 .offset = offsetof(struct xfrm_flow_keys, gre),
4372         },
4373         {
4374                 .key_id = FLOW_DISSECTOR_KEY_IP,
4375                 .offset = offsetof(struct xfrm_flow_keys, ip),
4376         },
4377         {
4378                 .key_id = FLOW_DISSECTOR_KEY_ICMP,
4379                 .offset = offsetof(struct xfrm_flow_keys, icmp),
4380         },
4381 };
4382
4383 void __init xfrm_init(void)
4384 {
4385         skb_flow_dissector_init(&xfrm_session_dissector,
4386                                 xfrm_flow_dissector_keys,
4387                                 ARRAY_SIZE(xfrm_flow_dissector_keys));
4388
4389         register_pernet_subsys(&xfrm_net_ops);
4390         xfrm_dev_init();
4391         xfrm_input_init();
4392
4393 #ifdef CONFIG_XFRM_ESPINTCP
4394         espintcp_init();
4395 #endif
4396
4397         register_xfrm_state_bpf();
4398         xfrm_nat_keepalive_init(AF_INET);
4399 }
4400
4401 #ifdef CONFIG_AUDITSYSCALL
4402 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp,
4403                                          struct audit_buffer *audit_buf)
4404 {
4405         struct xfrm_sec_ctx *ctx = xp->security;
4406         struct xfrm_selector *sel = &xp->selector;
4407
4408         if (ctx)
4409                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
4410                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
4411
4412         switch (sel->family) {
4413         case AF_INET:
4414                 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4);
4415                 if (sel->prefixlen_s != 32)
4416                         audit_log_format(audit_buf, " src_prefixlen=%d",
4417                                          sel->prefixlen_s);
4418                 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4);
4419                 if (sel->prefixlen_d != 32)
4420                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4421                                          sel->prefixlen_d);
4422                 break;
4423         case AF_INET6:
4424                 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6);
4425                 if (sel->prefixlen_s != 128)
4426                         audit_log_format(audit_buf, " src_prefixlen=%d",
4427                                          sel->prefixlen_s);
4428                 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6);
4429                 if (sel->prefixlen_d != 128)
4430                         audit_log_format(audit_buf, " dst_prefixlen=%d",
4431                                          sel->prefixlen_d);
4432                 break;
4433         }
4434 }
4435
4436 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid)
4437 {
4438         struct audit_buffer *audit_buf;
4439
4440         audit_buf = xfrm_audit_start("SPD-add");
4441         if (audit_buf == NULL)
4442                 return;
4443         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4444         audit_log_format(audit_buf, " res=%u", result);
4445         xfrm_audit_common_policyinfo(xp, audit_buf);
4446         audit_log_end(audit_buf);
4447 }
4448 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add);
4449
4450 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result,
4451                               bool task_valid)
4452 {
4453         struct audit_buffer *audit_buf;
4454
4455         audit_buf = xfrm_audit_start("SPD-delete");
4456         if (audit_buf == NULL)
4457                 return;
4458         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
4459         audit_log_format(audit_buf, " res=%u", result);
4460         xfrm_audit_common_policyinfo(xp, audit_buf);
4461         audit_log_end(audit_buf);
4462 }
4463 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete);
4464 #endif
4465
4466 #ifdef CONFIG_XFRM_MIGRATE
4467 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel,
4468                                                     u8 dir, u8 type, struct net *net, u32 if_id)
4469 {
4470         struct xfrm_policy *pol;
4471         struct flowi fl;
4472
4473         memset(&fl, 0, sizeof(fl));
4474
4475         fl.flowi_proto = sel->proto;
4476
4477         switch (sel->family) {
4478         case AF_INET:
4479                 fl.u.ip4.saddr = sel->saddr.a4;
4480                 fl.u.ip4.daddr = sel->daddr.a4;
4481                 if (sel->proto == IPSEC_ULPROTO_ANY)
4482                         break;
4483                 fl.u.flowi4_oif = sel->ifindex;
4484                 fl.u.ip4.fl4_sport = sel->sport;
4485                 fl.u.ip4.fl4_dport = sel->dport;
4486                 break;
4487         case AF_INET6:
4488                 fl.u.ip6.saddr = sel->saddr.in6;
4489                 fl.u.ip6.daddr = sel->daddr.in6;
4490                 if (sel->proto == IPSEC_ULPROTO_ANY)
4491                         break;
4492                 fl.u.flowi6_oif = sel->ifindex;
4493                 fl.u.ip6.fl4_sport = sel->sport;
4494                 fl.u.ip6.fl4_dport = sel->dport;
4495                 break;
4496         default:
4497                 return ERR_PTR(-EAFNOSUPPORT);
4498         }
4499
4500         rcu_read_lock();
4501
4502         pol = xfrm_policy_lookup_bytype(net, type, &fl, sel->family, dir, if_id);
4503         if (IS_ERR_OR_NULL(pol))
4504                 goto out_unlock;
4505
4506         if (!xfrm_pol_hold_rcu(pol))
4507                 pol = NULL;
4508 out_unlock:
4509         rcu_read_unlock();
4510         return pol;
4511 }
4512
4513 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t)
4514 {
4515         int match = 0;
4516
4517         if (t->mode == m->mode && t->id.proto == m->proto &&
4518             (m->reqid == 0 || t->reqid == m->reqid)) {
4519                 switch (t->mode) {
4520                 case XFRM_MODE_TUNNEL:
4521                 case XFRM_MODE_BEET:
4522                         if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr,
4523                                             m->old_family) &&
4524                             xfrm_addr_equal(&t->saddr, &m->old_saddr,
4525                                             m->old_family)) {
4526                                 match = 1;
4527                         }
4528                         break;
4529                 case XFRM_MODE_TRANSPORT:
4530                         /* in case of transport mode, template does not store
4531                            any IP addresses, hence we just compare mode and
4532                            protocol */
4533                         match = 1;
4534                         break;
4535                 default:
4536                         break;
4537                 }
4538         }
4539         return match;
4540 }
4541
4542 /* update endpoint address(es) of template(s) */
4543 static int xfrm_policy_migrate(struct xfrm_policy *pol,
4544                                struct xfrm_migrate *m, int num_migrate,
4545                                struct netlink_ext_ack *extack)
4546 {
4547         struct xfrm_migrate *mp;
4548         int i, j, n = 0;
4549
4550         write_lock_bh(&pol->lock);
4551         if (unlikely(pol->walk.dead)) {
4552                 /* target policy has been deleted */
4553                 NL_SET_ERR_MSG(extack, "Target policy not found");
4554                 write_unlock_bh(&pol->lock);
4555                 return -ENOENT;
4556         }
4557
4558         for (i = 0; i < pol->xfrm_nr; i++) {
4559                 for (j = 0, mp = m; j < num_migrate; j++, mp++) {
4560                         if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i]))
4561                                 continue;
4562                         n++;
4563                         if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL &&
4564                             pol->xfrm_vec[i].mode != XFRM_MODE_BEET)
4565                                 continue;
4566                         /* update endpoints */
4567                         memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr,
4568                                sizeof(pol->xfrm_vec[i].id.daddr));
4569                         memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr,
4570                                sizeof(pol->xfrm_vec[i].saddr));
4571                         pol->xfrm_vec[i].encap_family = mp->new_family;
4572                         /* flush bundles */
4573                         atomic_inc(&pol->genid);
4574                 }
4575         }
4576
4577         write_unlock_bh(&pol->lock);
4578
4579         if (!n)
4580                 return -ENODATA;
4581
4582         return 0;
4583 }
4584
4585 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate,
4586                               struct netlink_ext_ack *extack)
4587 {
4588         int i, j;
4589
4590         if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH) {
4591                 NL_SET_ERR_MSG(extack, "Invalid number of SAs to migrate, must be 0 < num <= XFRM_MAX_DEPTH (6)");
4592                 return -EINVAL;
4593         }
4594
4595         for (i = 0; i < num_migrate; i++) {
4596                 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) ||
4597                     xfrm_addr_any(&m[i].new_saddr, m[i].new_family)) {
4598                         NL_SET_ERR_MSG(extack, "Addresses in the MIGRATE attribute's list cannot be null");
4599                         return -EINVAL;
4600                 }
4601
4602                 /* check if there is any duplicated entry */
4603                 for (j = i + 1; j < num_migrate; j++) {
4604                         if (!memcmp(&m[i].old_daddr, &m[j].old_daddr,
4605                                     sizeof(m[i].old_daddr)) &&
4606                             !memcmp(&m[i].old_saddr, &m[j].old_saddr,
4607                                     sizeof(m[i].old_saddr)) &&
4608                             m[i].proto == m[j].proto &&
4609                             m[i].mode == m[j].mode &&
4610                             m[i].reqid == m[j].reqid &&
4611                             m[i].old_family == m[j].old_family) {
4612                                 NL_SET_ERR_MSG(extack, "Entries in the MIGRATE attribute's list must be unique");
4613                                 return -EINVAL;
4614                         }
4615                 }
4616         }
4617
4618         return 0;
4619 }
4620
4621 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
4622                  struct xfrm_migrate *m, int num_migrate,
4623                  struct xfrm_kmaddress *k, struct net *net,
4624                  struct xfrm_encap_tmpl *encap, u32 if_id,
4625                  struct netlink_ext_ack *extack)
4626 {
4627         int i, err, nx_cur = 0, nx_new = 0;
4628         struct xfrm_policy *pol = NULL;
4629         struct xfrm_state *x, *xc;
4630         struct xfrm_state *x_cur[XFRM_MAX_DEPTH];
4631         struct xfrm_state *x_new[XFRM_MAX_DEPTH];
4632         struct xfrm_migrate *mp;
4633
4634         /* Stage 0 - sanity checks */
4635         err = xfrm_migrate_check(m, num_migrate, extack);
4636         if (err < 0)
4637                 goto out;
4638
4639         if (dir >= XFRM_POLICY_MAX) {
4640                 NL_SET_ERR_MSG(extack, "Invalid policy direction");
4641                 err = -EINVAL;
4642                 goto out;
4643         }
4644
4645         /* Stage 1 - find policy */
4646         pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id);
4647         if (IS_ERR_OR_NULL(pol)) {
4648                 NL_SET_ERR_MSG(extack, "Target policy not found");
4649                 err = IS_ERR(pol) ? PTR_ERR(pol) : -ENOENT;
4650                 goto out;
4651         }
4652
4653         /* Stage 2 - find and update state(s) */
4654         for (i = 0, mp = m; i < num_migrate; i++, mp++) {
4655                 if ((x = xfrm_migrate_state_find(mp, net, if_id))) {
4656                         x_cur[nx_cur] = x;
4657                         nx_cur++;
4658                         xc = xfrm_state_migrate(x, mp, encap);
4659                         if (xc) {
4660                                 x_new[nx_new] = xc;
4661                                 nx_new++;
4662                         } else {
4663                                 err = -ENODATA;
4664                                 goto restore_state;
4665                         }
4666                 }
4667         }
4668
4669         /* Stage 3 - update policy */
4670         err = xfrm_policy_migrate(pol, m, num_migrate, extack);
4671         if (err < 0)
4672                 goto restore_state;
4673
4674         /* Stage 4 - delete old state(s) */
4675         if (nx_cur) {
4676                 xfrm_states_put(x_cur, nx_cur);
4677                 xfrm_states_delete(x_cur, nx_cur);
4678         }
4679
4680         /* Stage 5 - announce */
4681         km_migrate(sel, dir, type, m, num_migrate, k, encap);
4682
4683         xfrm_pol_put(pol);
4684
4685         return 0;
4686 out:
4687         return err;
4688
4689 restore_state:
4690         if (pol)
4691                 xfrm_pol_put(pol);
4692         if (nx_cur)
4693                 xfrm_states_put(x_cur, nx_cur);
4694         if (nx_new)
4695                 xfrm_states_delete(x_new, nx_new);
4696
4697         return err;
4698 }
4699 EXPORT_SYMBOL(xfrm_migrate);
4700 #endif