1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * xfrm_policy.c 4 * 5 * Changes: 6 * Mitsuru KANDA @USAGI 7 * Kazunori MIYAZAWA @USAGI 8 * Kunihiro Ishiguro <kunihiro@ipinfusion.com> 9 * IPv6 support 10 * Kazunori MIYAZAWA @USAGI 11 * YOSHIFUJI Hideaki 12 * Split up af-specific portion 13 * Derek Atkins <derek@ihtfp.com> Add the post_input processor 14 * 15 */ 16 17 #include <linux/err.h> 18 #include <linux/slab.h> 19 #include <linux/kmod.h> 20 #include <linux/list.h> 21 #include <linux/spinlock.h> 22 #include <linux/workqueue.h> 23 #include <linux/notifier.h> 24 #include <linux/netdevice.h> 25 #include <linux/netfilter.h> 26 #include <linux/module.h> 27 #include <linux/cache.h> 28 #include <linux/cpu.h> 29 #include <linux/audit.h> 30 #include <linux/rhashtable.h> 31 #include <linux/if_tunnel.h> 32 #include <net/dst.h> 33 #include <net/flow.h> 34 #include <net/inet_ecn.h> 35 #include <net/xfrm.h> 36 #include <net/ip.h> 37 #include <net/gre.h> 38 #if IS_ENABLED(CONFIG_IPV6_MIP6) 39 #include <net/mip6.h> 40 #endif 41 #ifdef CONFIG_XFRM_STATISTICS 42 #include <net/snmp.h> 43 #endif 44 #ifdef CONFIG_XFRM_ESPINTCP 45 #include <net/espintcp.h> 46 #endif 47 48 #include "xfrm_hash.h" 49 50 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10)) 51 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ)) 52 #define XFRM_MAX_QUEUE_LEN 100 53 54 struct xfrm_flo { 55 struct dst_entry *dst_orig; 56 u8 flags; 57 }; 58 59 /* prefixes smaller than this are stored in lists, not trees. */ 60 #define INEXACT_PREFIXLEN_IPV4 16 61 #define INEXACT_PREFIXLEN_IPV6 48 62 63 struct xfrm_pol_inexact_node { 64 struct rb_node node; 65 union { 66 xfrm_address_t addr; 67 struct rcu_head rcu; 68 }; 69 u8 prefixlen; 70 71 struct rb_root root; 72 73 /* the policies matching this node, can be empty list */ 74 struct hlist_head hhead; 75 }; 76 77 /* xfrm inexact policy search tree: 78 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id); 79 * | 80 * +---- root_d: sorted by daddr:prefix 81 * | | 82 * | xfrm_pol_inexact_node 83 * | | 84 * | +- root: sorted by saddr/prefix 85 * | | | 86 * | | xfrm_pol_inexact_node 87 * | | | 88 * | | + root: unused 89 * | | | 90 * | | + hhead: saddr:daddr policies 91 * | | 92 * | +- coarse policies and all any:daddr policies 93 * | 94 * +---- root_s: sorted by saddr:prefix 95 * | | 96 * | xfrm_pol_inexact_node 97 * | | 98 * | + root: unused 99 * | | 100 * | + hhead: saddr:any policies 101 * | 102 * +---- coarse policies and all any:any policies 103 * 104 * Lookups return four candidate lists: 105 * 1. any:any list from top-level xfrm_pol_inexact_bin 106 * 2. any:daddr list from daddr tree 107 * 3. saddr:daddr list from 2nd level daddr tree 108 * 4. saddr:any list from saddr tree 109 * 110 * This result set then needs to be searched for the policy with 111 * the lowest priority. If two results have same prio, youngest one wins. 112 */ 113 114 struct xfrm_pol_inexact_key { 115 possible_net_t net; 116 u32 if_id; 117 u16 family; 118 u8 dir, type; 119 }; 120 121 struct xfrm_pol_inexact_bin { 122 struct xfrm_pol_inexact_key k; 123 struct rhash_head head; 124 /* list containing '*:*' policies */ 125 struct hlist_head hhead; 126 127 seqcount_spinlock_t count; 128 /* tree sorted by daddr/prefix */ 129 struct rb_root root_d; 130 131 /* tree sorted by saddr/prefix */ 132 struct rb_root root_s; 133 134 /* slow path below */ 135 struct list_head inexact_bins; 136 struct rcu_head rcu; 137 }; 138 139 enum xfrm_pol_inexact_candidate_type { 140 XFRM_POL_CAND_BOTH, 141 XFRM_POL_CAND_SADDR, 142 XFRM_POL_CAND_DADDR, 143 XFRM_POL_CAND_ANY, 144 145 XFRM_POL_CAND_MAX, 146 }; 147 148 struct xfrm_pol_inexact_candidates { 149 struct hlist_head *res[XFRM_POL_CAND_MAX]; 150 }; 151 152 static DEFINE_SPINLOCK(xfrm_if_cb_lock); 153 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly; 154 155 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock); 156 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1] 157 __read_mostly; 158 159 static struct kmem_cache *xfrm_dst_cache __ro_after_init; 160 161 static struct rhashtable xfrm_policy_inexact_table; 162 static const struct rhashtable_params xfrm_pol_inexact_params; 163 164 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr); 165 static int stale_bundle(struct dst_entry *dst); 166 static int xfrm_bundle_ok(struct xfrm_dst *xdst); 167 static void xfrm_policy_queue_process(struct timer_list *t); 168 169 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir); 170 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol, 171 int dir); 172 173 static struct xfrm_pol_inexact_bin * 174 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir, 175 u32 if_id); 176 177 static struct xfrm_pol_inexact_bin * 178 xfrm_policy_inexact_lookup_rcu(struct net *net, 179 u8 type, u16 family, u8 dir, u32 if_id); 180 static struct xfrm_policy * 181 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy, 182 bool excl); 183 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain, 184 struct xfrm_policy *policy); 185 186 static bool 187 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand, 188 struct xfrm_pol_inexact_bin *b, 189 const xfrm_address_t *saddr, 190 const xfrm_address_t *daddr); 191 192 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy) 193 { 194 return refcount_inc_not_zero(&policy->refcnt); 195 } 196 197 static inline bool 198 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl) 199 { 200 const struct flowi4 *fl4 = &fl->u.ip4; 201 202 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) && 203 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) && 204 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) && 205 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) && 206 (fl4->flowi4_proto == sel->proto || !sel->proto) && 207 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex); 208 } 209 210 static inline bool 211 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl) 212 { 213 const struct flowi6 *fl6 = &fl->u.ip6; 214 215 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) && 216 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) && 217 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) && 218 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) && 219 (fl6->flowi6_proto == sel->proto || !sel->proto) && 220 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex); 221 } 222 223 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl, 224 unsigned short family) 225 { 226 switch (family) { 227 case AF_INET: 228 return __xfrm4_selector_match(sel, fl); 229 case AF_INET6: 230 return __xfrm6_selector_match(sel, fl); 231 } 232 return false; 233 } 234 235 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family) 236 { 237 const struct xfrm_policy_afinfo *afinfo; 238 239 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo))) 240 return NULL; 241 rcu_read_lock(); 242 afinfo = rcu_dereference(xfrm_policy_afinfo[family]); 243 if (unlikely(!afinfo)) 244 rcu_read_unlock(); 245 return afinfo; 246 } 247 248 /* Called with rcu_read_lock(). */ 249 static const struct xfrm_if_cb *xfrm_if_get_cb(void) 250 { 251 return rcu_dereference(xfrm_if_cb); 252 } 253 254 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif, 255 const xfrm_address_t *saddr, 256 const xfrm_address_t *daddr, 257 int family, u32 mark) 258 { 259 const struct xfrm_policy_afinfo *afinfo; 260 struct dst_entry *dst; 261 262 afinfo = xfrm_policy_get_afinfo(family); 263 if (unlikely(afinfo == NULL)) 264 return ERR_PTR(-EAFNOSUPPORT); 265 266 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark); 267 268 rcu_read_unlock(); 269 270 return dst; 271 } 272 EXPORT_SYMBOL(__xfrm_dst_lookup); 273 274 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, 275 int tos, int oif, 276 xfrm_address_t *prev_saddr, 277 xfrm_address_t *prev_daddr, 278 int family, u32 mark) 279 { 280 struct net *net = xs_net(x); 281 xfrm_address_t *saddr = &x->props.saddr; 282 xfrm_address_t *daddr = &x->id.daddr; 283 struct dst_entry *dst; 284 285 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) { 286 saddr = x->coaddr; 287 daddr = prev_daddr; 288 } 289 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) { 290 saddr = prev_saddr; 291 daddr = x->coaddr; 292 } 293 294 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark); 295 296 if (!IS_ERR(dst)) { 297 if (prev_saddr != saddr) 298 memcpy(prev_saddr, saddr, sizeof(*prev_saddr)); 299 if (prev_daddr != daddr) 300 memcpy(prev_daddr, daddr, sizeof(*prev_daddr)); 301 } 302 303 return dst; 304 } 305 306 static inline unsigned long make_jiffies(long secs) 307 { 308 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ) 309 return MAX_SCHEDULE_TIMEOUT-1; 310 else 311 return secs*HZ; 312 } 313 314 static void xfrm_policy_timer(struct timer_list *t) 315 { 316 struct xfrm_policy *xp = from_timer(xp, t, timer); 317 time64_t now = ktime_get_real_seconds(); 318 time64_t next = TIME64_MAX; 319 int warn = 0; 320 int dir; 321 322 read_lock(&xp->lock); 323 324 if (unlikely(xp->walk.dead)) 325 goto out; 326 327 dir = xfrm_policy_id2dir(xp->index); 328 329 if (xp->lft.hard_add_expires_seconds) { 330 time64_t tmo = xp->lft.hard_add_expires_seconds + 331 xp->curlft.add_time - now; 332 if (tmo <= 0) 333 goto expired; 334 if (tmo < next) 335 next = tmo; 336 } 337 if (xp->lft.hard_use_expires_seconds) { 338 time64_t tmo = xp->lft.hard_use_expires_seconds + 339 (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now; 340 if (tmo <= 0) 341 goto expired; 342 if (tmo < next) 343 next = tmo; 344 } 345 if (xp->lft.soft_add_expires_seconds) { 346 time64_t tmo = xp->lft.soft_add_expires_seconds + 347 xp->curlft.add_time - now; 348 if (tmo <= 0) { 349 warn = 1; 350 tmo = XFRM_KM_TIMEOUT; 351 } 352 if (tmo < next) 353 next = tmo; 354 } 355 if (xp->lft.soft_use_expires_seconds) { 356 time64_t tmo = xp->lft.soft_use_expires_seconds + 357 (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now; 358 if (tmo <= 0) { 359 warn = 1; 360 tmo = XFRM_KM_TIMEOUT; 361 } 362 if (tmo < next) 363 next = tmo; 364 } 365 366 if (warn) 367 km_policy_expired(xp, dir, 0, 0); 368 if (next != TIME64_MAX && 369 !mod_timer(&xp->timer, jiffies + make_jiffies(next))) 370 xfrm_pol_hold(xp); 371 372 out: 373 read_unlock(&xp->lock); 374 xfrm_pol_put(xp); 375 return; 376 377 expired: 378 read_unlock(&xp->lock); 379 if (!xfrm_policy_delete(xp, dir)) 380 km_policy_expired(xp, dir, 1, 0); 381 xfrm_pol_put(xp); 382 } 383 384 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2 385 * SPD calls. 386 */ 387 388 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp) 389 { 390 struct xfrm_policy *policy; 391 392 policy = kzalloc(sizeof(struct xfrm_policy), gfp); 393 394 if (policy) { 395 write_pnet(&policy->xp_net, net); 396 INIT_LIST_HEAD(&policy->walk.all); 397 INIT_HLIST_NODE(&policy->bydst_inexact_list); 398 INIT_HLIST_NODE(&policy->bydst); 399 INIT_HLIST_NODE(&policy->byidx); 400 rwlock_init(&policy->lock); 401 refcount_set(&policy->refcnt, 1); 402 skb_queue_head_init(&policy->polq.hold_queue); 403 timer_setup(&policy->timer, xfrm_policy_timer, 0); 404 timer_setup(&policy->polq.hold_timer, 405 xfrm_policy_queue_process, 0); 406 } 407 return policy; 408 } 409 EXPORT_SYMBOL(xfrm_policy_alloc); 410 411 static void xfrm_policy_destroy_rcu(struct rcu_head *head) 412 { 413 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu); 414 415 security_xfrm_policy_free(policy->security); 416 kfree(policy); 417 } 418 419 /* Destroy xfrm_policy: descendant resources must be released to this moment. */ 420 421 void xfrm_policy_destroy(struct xfrm_policy *policy) 422 { 423 BUG_ON(!policy->walk.dead); 424 425 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer)) 426 BUG(); 427 428 xfrm_dev_policy_free(policy); 429 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu); 430 } 431 EXPORT_SYMBOL(xfrm_policy_destroy); 432 433 /* Rule must be locked. Release descendant resources, announce 434 * entry dead. The rule must be unlinked from lists to the moment. 435 */ 436 437 static void xfrm_policy_kill(struct xfrm_policy *policy) 438 { 439 xfrm_dev_policy_delete(policy); 440 441 write_lock_bh(&policy->lock); 442 policy->walk.dead = 1; 443 write_unlock_bh(&policy->lock); 444 445 atomic_inc(&policy->genid); 446 447 if (del_timer(&policy->polq.hold_timer)) 448 xfrm_pol_put(policy); 449 skb_queue_purge(&policy->polq.hold_queue); 450 451 if (del_timer(&policy->timer)) 452 xfrm_pol_put(policy); 453 454 xfrm_pol_put(policy); 455 } 456 457 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024; 458 459 static inline unsigned int idx_hash(struct net *net, u32 index) 460 { 461 return __idx_hash(index, net->xfrm.policy_idx_hmask); 462 } 463 464 /* calculate policy hash thresholds */ 465 static void __get_hash_thresh(struct net *net, 466 unsigned short family, int dir, 467 u8 *dbits, u8 *sbits) 468 { 469 switch (family) { 470 case AF_INET: 471 *dbits = net->xfrm.policy_bydst[dir].dbits4; 472 *sbits = net->xfrm.policy_bydst[dir].sbits4; 473 break; 474 475 case AF_INET6: 476 *dbits = net->xfrm.policy_bydst[dir].dbits6; 477 *sbits = net->xfrm.policy_bydst[dir].sbits6; 478 break; 479 480 default: 481 *dbits = 0; 482 *sbits = 0; 483 } 484 } 485 486 static struct hlist_head *policy_hash_bysel(struct net *net, 487 const struct xfrm_selector *sel, 488 unsigned short family, int dir) 489 { 490 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 491 unsigned int hash; 492 u8 dbits; 493 u8 sbits; 494 495 __get_hash_thresh(net, family, dir, &dbits, &sbits); 496 hash = __sel_hash(sel, family, hmask, dbits, sbits); 497 498 if (hash == hmask + 1) 499 return NULL; 500 501 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table, 502 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash; 503 } 504 505 static struct hlist_head *policy_hash_direct(struct net *net, 506 const xfrm_address_t *daddr, 507 const xfrm_address_t *saddr, 508 unsigned short family, int dir) 509 { 510 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 511 unsigned int hash; 512 u8 dbits; 513 u8 sbits; 514 515 __get_hash_thresh(net, family, dir, &dbits, &sbits); 516 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits); 517 518 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table, 519 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash; 520 } 521 522 static void xfrm_dst_hash_transfer(struct net *net, 523 struct hlist_head *list, 524 struct hlist_head *ndsttable, 525 unsigned int nhashmask, 526 int dir) 527 { 528 struct hlist_node *tmp, *entry0 = NULL; 529 struct xfrm_policy *pol; 530 unsigned int h0 = 0; 531 u8 dbits; 532 u8 sbits; 533 534 redo: 535 hlist_for_each_entry_safe(pol, tmp, list, bydst) { 536 unsigned int h; 537 538 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits); 539 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr, 540 pol->family, nhashmask, dbits, sbits); 541 if (!entry0 || pol->xdo.type == XFRM_DEV_OFFLOAD_PACKET) { 542 hlist_del_rcu(&pol->bydst); 543 hlist_add_head_rcu(&pol->bydst, ndsttable + h); 544 h0 = h; 545 } else { 546 if (h != h0) 547 continue; 548 hlist_del_rcu(&pol->bydst); 549 hlist_add_behind_rcu(&pol->bydst, entry0); 550 } 551 entry0 = &pol->bydst; 552 } 553 if (!hlist_empty(list)) { 554 entry0 = NULL; 555 goto redo; 556 } 557 } 558 559 static void xfrm_idx_hash_transfer(struct hlist_head *list, 560 struct hlist_head *nidxtable, 561 unsigned int nhashmask) 562 { 563 struct hlist_node *tmp; 564 struct xfrm_policy *pol; 565 566 hlist_for_each_entry_safe(pol, tmp, list, byidx) { 567 unsigned int h; 568 569 h = __idx_hash(pol->index, nhashmask); 570 hlist_add_head(&pol->byidx, nidxtable+h); 571 } 572 } 573 574 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask) 575 { 576 return ((old_hmask + 1) << 1) - 1; 577 } 578 579 static void xfrm_bydst_resize(struct net *net, int dir) 580 { 581 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 582 unsigned int nhashmask = xfrm_new_hash_mask(hmask); 583 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head); 584 struct hlist_head *ndst = xfrm_hash_alloc(nsize); 585 struct hlist_head *odst; 586 int i; 587 588 if (!ndst) 589 return; 590 591 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 592 write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation); 593 594 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table, 595 lockdep_is_held(&net->xfrm.xfrm_policy_lock)); 596 597 for (i = hmask; i >= 0; i--) 598 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir); 599 600 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst); 601 net->xfrm.policy_bydst[dir].hmask = nhashmask; 602 603 write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation); 604 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 605 606 synchronize_rcu(); 607 608 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head)); 609 } 610 611 static void xfrm_byidx_resize(struct net *net) 612 { 613 unsigned int hmask = net->xfrm.policy_idx_hmask; 614 unsigned int nhashmask = xfrm_new_hash_mask(hmask); 615 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head); 616 struct hlist_head *oidx = net->xfrm.policy_byidx; 617 struct hlist_head *nidx = xfrm_hash_alloc(nsize); 618 int i; 619 620 if (!nidx) 621 return; 622 623 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 624 625 for (i = hmask; i >= 0; i--) 626 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask); 627 628 net->xfrm.policy_byidx = nidx; 629 net->xfrm.policy_idx_hmask = nhashmask; 630 631 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 632 633 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head)); 634 } 635 636 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total) 637 { 638 unsigned int cnt = net->xfrm.policy_count[dir]; 639 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 640 641 if (total) 642 *total += cnt; 643 644 if ((hmask + 1) < xfrm_policy_hashmax && 645 cnt > hmask) 646 return 1; 647 648 return 0; 649 } 650 651 static inline int xfrm_byidx_should_resize(struct net *net, int total) 652 { 653 unsigned int hmask = net->xfrm.policy_idx_hmask; 654 655 if ((hmask + 1) < xfrm_policy_hashmax && 656 total > hmask) 657 return 1; 658 659 return 0; 660 } 661 662 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si) 663 { 664 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN]; 665 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT]; 666 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD]; 667 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX]; 668 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX]; 669 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX]; 670 si->spdhcnt = net->xfrm.policy_idx_hmask; 671 si->spdhmcnt = xfrm_policy_hashmax; 672 } 673 EXPORT_SYMBOL(xfrm_spd_getinfo); 674 675 static DEFINE_MUTEX(hash_resize_mutex); 676 static void xfrm_hash_resize(struct work_struct *work) 677 { 678 struct net *net = container_of(work, struct net, xfrm.policy_hash_work); 679 int dir, total; 680 681 mutex_lock(&hash_resize_mutex); 682 683 total = 0; 684 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 685 if (xfrm_bydst_should_resize(net, dir, &total)) 686 xfrm_bydst_resize(net, dir); 687 } 688 if (xfrm_byidx_should_resize(net, total)) 689 xfrm_byidx_resize(net); 690 691 mutex_unlock(&hash_resize_mutex); 692 } 693 694 /* Make sure *pol can be inserted into fastbin. 695 * Useful to check that later insert requests will be successful 696 * (provided xfrm_policy_lock is held throughout). 697 */ 698 static struct xfrm_pol_inexact_bin * 699 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir) 700 { 701 struct xfrm_pol_inexact_bin *bin, *prev; 702 struct xfrm_pol_inexact_key k = { 703 .family = pol->family, 704 .type = pol->type, 705 .dir = dir, 706 .if_id = pol->if_id, 707 }; 708 struct net *net = xp_net(pol); 709 710 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 711 712 write_pnet(&k.net, net); 713 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k, 714 xfrm_pol_inexact_params); 715 if (bin) 716 return bin; 717 718 bin = kzalloc(sizeof(*bin), GFP_ATOMIC); 719 if (!bin) 720 return NULL; 721 722 bin->k = k; 723 INIT_HLIST_HEAD(&bin->hhead); 724 bin->root_d = RB_ROOT; 725 bin->root_s = RB_ROOT; 726 seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock); 727 728 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table, 729 &bin->k, &bin->head, 730 xfrm_pol_inexact_params); 731 if (!prev) { 732 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins); 733 return bin; 734 } 735 736 kfree(bin); 737 738 return IS_ERR(prev) ? NULL : prev; 739 } 740 741 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr, 742 int family, u8 prefixlen) 743 { 744 if (xfrm_addr_any(addr, family)) 745 return true; 746 747 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6) 748 return true; 749 750 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4) 751 return true; 752 753 return false; 754 } 755 756 static bool 757 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy) 758 { 759 const xfrm_address_t *addr; 760 bool saddr_any, daddr_any; 761 u8 prefixlen; 762 763 addr = &policy->selector.saddr; 764 prefixlen = policy->selector.prefixlen_s; 765 766 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr, 767 policy->family, 768 prefixlen); 769 addr = &policy->selector.daddr; 770 prefixlen = policy->selector.prefixlen_d; 771 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr, 772 policy->family, 773 prefixlen); 774 return saddr_any && daddr_any; 775 } 776 777 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node, 778 const xfrm_address_t *addr, u8 prefixlen) 779 { 780 node->addr = *addr; 781 node->prefixlen = prefixlen; 782 } 783 784 static struct xfrm_pol_inexact_node * 785 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen) 786 { 787 struct xfrm_pol_inexact_node *node; 788 789 node = kzalloc(sizeof(*node), GFP_ATOMIC); 790 if (node) 791 xfrm_pol_inexact_node_init(node, addr, prefixlen); 792 793 return node; 794 } 795 796 static int xfrm_policy_addr_delta(const xfrm_address_t *a, 797 const xfrm_address_t *b, 798 u8 prefixlen, u16 family) 799 { 800 u32 ma, mb, mask; 801 unsigned int pdw, pbi; 802 int delta = 0; 803 804 switch (family) { 805 case AF_INET: 806 if (prefixlen == 0) 807 return 0; 808 mask = ~0U << (32 - prefixlen); 809 ma = ntohl(a->a4) & mask; 810 mb = ntohl(b->a4) & mask; 811 if (ma < mb) 812 delta = -1; 813 else if (ma > mb) 814 delta = 1; 815 break; 816 case AF_INET6: 817 pdw = prefixlen >> 5; 818 pbi = prefixlen & 0x1f; 819 820 if (pdw) { 821 delta = memcmp(a->a6, b->a6, pdw << 2); 822 if (delta) 823 return delta; 824 } 825 if (pbi) { 826 mask = ~0U << (32 - pbi); 827 ma = ntohl(a->a6[pdw]) & mask; 828 mb = ntohl(b->a6[pdw]) & mask; 829 if (ma < mb) 830 delta = -1; 831 else if (ma > mb) 832 delta = 1; 833 } 834 break; 835 default: 836 break; 837 } 838 839 return delta; 840 } 841 842 static void xfrm_policy_inexact_list_reinsert(struct net *net, 843 struct xfrm_pol_inexact_node *n, 844 u16 family) 845 { 846 unsigned int matched_s, matched_d; 847 struct xfrm_policy *policy, *p; 848 849 matched_s = 0; 850 matched_d = 0; 851 852 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) { 853 struct hlist_node *newpos = NULL; 854 bool matches_s, matches_d; 855 856 if (policy->walk.dead || !policy->bydst_reinsert) 857 continue; 858 859 WARN_ON_ONCE(policy->family != family); 860 861 policy->bydst_reinsert = false; 862 hlist_for_each_entry(p, &n->hhead, bydst) { 863 if (policy->priority > p->priority) 864 newpos = &p->bydst; 865 else if (policy->priority == p->priority && 866 policy->pos > p->pos) 867 newpos = &p->bydst; 868 else 869 break; 870 } 871 872 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET) 873 hlist_add_behind_rcu(&policy->bydst, newpos); 874 else 875 hlist_add_head_rcu(&policy->bydst, &n->hhead); 876 877 /* paranoia checks follow. 878 * Check that the reinserted policy matches at least 879 * saddr or daddr for current node prefix. 880 * 881 * Matching both is fine, matching saddr in one policy 882 * (but not daddr) and then matching only daddr in another 883 * is a bug. 884 */ 885 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr, 886 &n->addr, 887 n->prefixlen, 888 family) == 0; 889 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr, 890 &n->addr, 891 n->prefixlen, 892 family) == 0; 893 if (matches_s && matches_d) 894 continue; 895 896 WARN_ON_ONCE(!matches_s && !matches_d); 897 if (matches_s) 898 matched_s++; 899 if (matches_d) 900 matched_d++; 901 WARN_ON_ONCE(matched_s && matched_d); 902 } 903 } 904 905 static void xfrm_policy_inexact_node_reinsert(struct net *net, 906 struct xfrm_pol_inexact_node *n, 907 struct rb_root *new, 908 u16 family) 909 { 910 struct xfrm_pol_inexact_node *node; 911 struct rb_node **p, *parent; 912 913 /* we should not have another subtree here */ 914 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root)); 915 restart: 916 parent = NULL; 917 p = &new->rb_node; 918 while (*p) { 919 u8 prefixlen; 920 int delta; 921 922 parent = *p; 923 node = rb_entry(*p, struct xfrm_pol_inexact_node, node); 924 925 prefixlen = min(node->prefixlen, n->prefixlen); 926 927 delta = xfrm_policy_addr_delta(&n->addr, &node->addr, 928 prefixlen, family); 929 if (delta < 0) { 930 p = &parent->rb_left; 931 } else if (delta > 0) { 932 p = &parent->rb_right; 933 } else { 934 bool same_prefixlen = node->prefixlen == n->prefixlen; 935 struct xfrm_policy *tmp; 936 937 hlist_for_each_entry(tmp, &n->hhead, bydst) { 938 tmp->bydst_reinsert = true; 939 hlist_del_rcu(&tmp->bydst); 940 } 941 942 node->prefixlen = prefixlen; 943 944 xfrm_policy_inexact_list_reinsert(net, node, family); 945 946 if (same_prefixlen) { 947 kfree_rcu(n, rcu); 948 return; 949 } 950 951 rb_erase(*p, new); 952 kfree_rcu(n, rcu); 953 n = node; 954 goto restart; 955 } 956 } 957 958 rb_link_node_rcu(&n->node, parent, p); 959 rb_insert_color(&n->node, new); 960 } 961 962 /* merge nodes v and n */ 963 static void xfrm_policy_inexact_node_merge(struct net *net, 964 struct xfrm_pol_inexact_node *v, 965 struct xfrm_pol_inexact_node *n, 966 u16 family) 967 { 968 struct xfrm_pol_inexact_node *node; 969 struct xfrm_policy *tmp; 970 struct rb_node *rnode; 971 972 /* To-be-merged node v has a subtree. 973 * 974 * Dismantle it and insert its nodes to n->root. 975 */ 976 while ((rnode = rb_first(&v->root)) != NULL) { 977 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node); 978 rb_erase(&node->node, &v->root); 979 xfrm_policy_inexact_node_reinsert(net, node, &n->root, 980 family); 981 } 982 983 hlist_for_each_entry(tmp, &v->hhead, bydst) { 984 tmp->bydst_reinsert = true; 985 hlist_del_rcu(&tmp->bydst); 986 } 987 988 xfrm_policy_inexact_list_reinsert(net, n, family); 989 } 990 991 static struct xfrm_pol_inexact_node * 992 xfrm_policy_inexact_insert_node(struct net *net, 993 struct rb_root *root, 994 xfrm_address_t *addr, 995 u16 family, u8 prefixlen, u8 dir) 996 { 997 struct xfrm_pol_inexact_node *cached = NULL; 998 struct rb_node **p, *parent = NULL; 999 struct xfrm_pol_inexact_node *node; 1000 1001 p = &root->rb_node; 1002 while (*p) { 1003 int delta; 1004 1005 parent = *p; 1006 node = rb_entry(*p, struct xfrm_pol_inexact_node, node); 1007 1008 delta = xfrm_policy_addr_delta(addr, &node->addr, 1009 node->prefixlen, 1010 family); 1011 if (delta == 0 && prefixlen >= node->prefixlen) { 1012 WARN_ON_ONCE(cached); /* ipsec policies got lost */ 1013 return node; 1014 } 1015 1016 if (delta < 0) 1017 p = &parent->rb_left; 1018 else 1019 p = &parent->rb_right; 1020 1021 if (prefixlen < node->prefixlen) { 1022 delta = xfrm_policy_addr_delta(addr, &node->addr, 1023 prefixlen, 1024 family); 1025 if (delta) 1026 continue; 1027 1028 /* This node is a subnet of the new prefix. It needs 1029 * to be removed and re-inserted with the smaller 1030 * prefix and all nodes that are now also covered 1031 * by the reduced prefixlen. 1032 */ 1033 rb_erase(&node->node, root); 1034 1035 if (!cached) { 1036 xfrm_pol_inexact_node_init(node, addr, 1037 prefixlen); 1038 cached = node; 1039 } else { 1040 /* This node also falls within the new 1041 * prefixlen. Merge the to-be-reinserted 1042 * node and this one. 1043 */ 1044 xfrm_policy_inexact_node_merge(net, node, 1045 cached, family); 1046 kfree_rcu(node, rcu); 1047 } 1048 1049 /* restart */ 1050 p = &root->rb_node; 1051 parent = NULL; 1052 } 1053 } 1054 1055 node = cached; 1056 if (!node) { 1057 node = xfrm_pol_inexact_node_alloc(addr, prefixlen); 1058 if (!node) 1059 return NULL; 1060 } 1061 1062 rb_link_node_rcu(&node->node, parent, p); 1063 rb_insert_color(&node->node, root); 1064 1065 return node; 1066 } 1067 1068 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm) 1069 { 1070 struct xfrm_pol_inexact_node *node; 1071 struct rb_node *rn = rb_first(r); 1072 1073 while (rn) { 1074 node = rb_entry(rn, struct xfrm_pol_inexact_node, node); 1075 1076 xfrm_policy_inexact_gc_tree(&node->root, rm); 1077 rn = rb_next(rn); 1078 1079 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) { 1080 WARN_ON_ONCE(rm); 1081 continue; 1082 } 1083 1084 rb_erase(&node->node, r); 1085 kfree_rcu(node, rcu); 1086 } 1087 } 1088 1089 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit) 1090 { 1091 write_seqcount_begin(&b->count); 1092 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit); 1093 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit); 1094 write_seqcount_end(&b->count); 1095 1096 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) || 1097 !hlist_empty(&b->hhead)) { 1098 WARN_ON_ONCE(net_exit); 1099 return; 1100 } 1101 1102 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head, 1103 xfrm_pol_inexact_params) == 0) { 1104 list_del(&b->inexact_bins); 1105 kfree_rcu(b, rcu); 1106 } 1107 } 1108 1109 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b) 1110 { 1111 struct net *net = read_pnet(&b->k.net); 1112 1113 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1114 __xfrm_policy_inexact_prune_bin(b, false); 1115 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1116 } 1117 1118 static void __xfrm_policy_inexact_flush(struct net *net) 1119 { 1120 struct xfrm_pol_inexact_bin *bin, *t; 1121 1122 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1123 1124 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins) 1125 __xfrm_policy_inexact_prune_bin(bin, false); 1126 } 1127 1128 static struct hlist_head * 1129 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin, 1130 struct xfrm_policy *policy, u8 dir) 1131 { 1132 struct xfrm_pol_inexact_node *n; 1133 struct net *net; 1134 1135 net = xp_net(policy); 1136 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1137 1138 if (xfrm_policy_inexact_insert_use_any_list(policy)) 1139 return &bin->hhead; 1140 1141 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr, 1142 policy->family, 1143 policy->selector.prefixlen_d)) { 1144 write_seqcount_begin(&bin->count); 1145 n = xfrm_policy_inexact_insert_node(net, 1146 &bin->root_s, 1147 &policy->selector.saddr, 1148 policy->family, 1149 policy->selector.prefixlen_s, 1150 dir); 1151 write_seqcount_end(&bin->count); 1152 if (!n) 1153 return NULL; 1154 1155 return &n->hhead; 1156 } 1157 1158 /* daddr is fixed */ 1159 write_seqcount_begin(&bin->count); 1160 n = xfrm_policy_inexact_insert_node(net, 1161 &bin->root_d, 1162 &policy->selector.daddr, 1163 policy->family, 1164 policy->selector.prefixlen_d, dir); 1165 write_seqcount_end(&bin->count); 1166 if (!n) 1167 return NULL; 1168 1169 /* saddr is wildcard */ 1170 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr, 1171 policy->family, 1172 policy->selector.prefixlen_s)) 1173 return &n->hhead; 1174 1175 write_seqcount_begin(&bin->count); 1176 n = xfrm_policy_inexact_insert_node(net, 1177 &n->root, 1178 &policy->selector.saddr, 1179 policy->family, 1180 policy->selector.prefixlen_s, dir); 1181 write_seqcount_end(&bin->count); 1182 if (!n) 1183 return NULL; 1184 1185 return &n->hhead; 1186 } 1187 1188 static struct xfrm_policy * 1189 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl) 1190 { 1191 struct xfrm_pol_inexact_bin *bin; 1192 struct xfrm_policy *delpol; 1193 struct hlist_head *chain; 1194 struct net *net; 1195 1196 bin = xfrm_policy_inexact_alloc_bin(policy, dir); 1197 if (!bin) 1198 return ERR_PTR(-ENOMEM); 1199 1200 net = xp_net(policy); 1201 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1202 1203 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir); 1204 if (!chain) { 1205 __xfrm_policy_inexact_prune_bin(bin, false); 1206 return ERR_PTR(-ENOMEM); 1207 } 1208 1209 delpol = xfrm_policy_insert_list(chain, policy, excl); 1210 if (delpol && excl) { 1211 __xfrm_policy_inexact_prune_bin(bin, false); 1212 return ERR_PTR(-EEXIST); 1213 } 1214 1215 chain = &net->xfrm.policy_inexact[dir]; 1216 xfrm_policy_insert_inexact_list(chain, policy); 1217 1218 if (delpol) 1219 __xfrm_policy_inexact_prune_bin(bin, false); 1220 1221 return delpol; 1222 } 1223 1224 static void xfrm_hash_rebuild(struct work_struct *work) 1225 { 1226 struct net *net = container_of(work, struct net, 1227 xfrm.policy_hthresh.work); 1228 unsigned int hmask; 1229 struct xfrm_policy *pol; 1230 struct xfrm_policy *policy; 1231 struct hlist_head *chain; 1232 struct hlist_head *odst; 1233 struct hlist_node *newpos; 1234 int i; 1235 int dir; 1236 unsigned seq; 1237 u8 lbits4, rbits4, lbits6, rbits6; 1238 1239 mutex_lock(&hash_resize_mutex); 1240 1241 /* read selector prefixlen thresholds */ 1242 do { 1243 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock); 1244 1245 lbits4 = net->xfrm.policy_hthresh.lbits4; 1246 rbits4 = net->xfrm.policy_hthresh.rbits4; 1247 lbits6 = net->xfrm.policy_hthresh.lbits6; 1248 rbits6 = net->xfrm.policy_hthresh.rbits6; 1249 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq)); 1250 1251 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1252 write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation); 1253 1254 /* make sure that we can insert the indirect policies again before 1255 * we start with destructive action. 1256 */ 1257 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) { 1258 struct xfrm_pol_inexact_bin *bin; 1259 u8 dbits, sbits; 1260 1261 if (policy->walk.dead) 1262 continue; 1263 1264 dir = xfrm_policy_id2dir(policy->index); 1265 if (dir >= XFRM_POLICY_MAX) 1266 continue; 1267 1268 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) { 1269 if (policy->family == AF_INET) { 1270 dbits = rbits4; 1271 sbits = lbits4; 1272 } else { 1273 dbits = rbits6; 1274 sbits = lbits6; 1275 } 1276 } else { 1277 if (policy->family == AF_INET) { 1278 dbits = lbits4; 1279 sbits = rbits4; 1280 } else { 1281 dbits = lbits6; 1282 sbits = rbits6; 1283 } 1284 } 1285 1286 if (policy->selector.prefixlen_d < dbits || 1287 policy->selector.prefixlen_s < sbits) 1288 continue; 1289 1290 bin = xfrm_policy_inexact_alloc_bin(policy, dir); 1291 if (!bin) 1292 goto out_unlock; 1293 1294 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir)) 1295 goto out_unlock; 1296 } 1297 1298 /* reset the bydst and inexact table in all directions */ 1299 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 1300 struct hlist_node *n; 1301 1302 hlist_for_each_entry_safe(policy, n, 1303 &net->xfrm.policy_inexact[dir], 1304 bydst_inexact_list) { 1305 hlist_del_rcu(&policy->bydst); 1306 hlist_del_init(&policy->bydst_inexact_list); 1307 } 1308 1309 hmask = net->xfrm.policy_bydst[dir].hmask; 1310 odst = net->xfrm.policy_bydst[dir].table; 1311 for (i = hmask; i >= 0; i--) { 1312 hlist_for_each_entry_safe(policy, n, odst + i, bydst) 1313 hlist_del_rcu(&policy->bydst); 1314 } 1315 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) { 1316 /* dir out => dst = remote, src = local */ 1317 net->xfrm.policy_bydst[dir].dbits4 = rbits4; 1318 net->xfrm.policy_bydst[dir].sbits4 = lbits4; 1319 net->xfrm.policy_bydst[dir].dbits6 = rbits6; 1320 net->xfrm.policy_bydst[dir].sbits6 = lbits6; 1321 } else { 1322 /* dir in/fwd => dst = local, src = remote */ 1323 net->xfrm.policy_bydst[dir].dbits4 = lbits4; 1324 net->xfrm.policy_bydst[dir].sbits4 = rbits4; 1325 net->xfrm.policy_bydst[dir].dbits6 = lbits6; 1326 net->xfrm.policy_bydst[dir].sbits6 = rbits6; 1327 } 1328 } 1329 1330 /* re-insert all policies by order of creation */ 1331 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) { 1332 if (policy->walk.dead) 1333 continue; 1334 dir = xfrm_policy_id2dir(policy->index); 1335 if (dir >= XFRM_POLICY_MAX) { 1336 /* skip socket policies */ 1337 continue; 1338 } 1339 newpos = NULL; 1340 chain = policy_hash_bysel(net, &policy->selector, 1341 policy->family, dir); 1342 1343 if (!chain) { 1344 void *p = xfrm_policy_inexact_insert(policy, dir, 0); 1345 1346 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p)); 1347 continue; 1348 } 1349 1350 hlist_for_each_entry(pol, chain, bydst) { 1351 if (policy->priority >= pol->priority) 1352 newpos = &pol->bydst; 1353 else 1354 break; 1355 } 1356 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET) 1357 hlist_add_behind_rcu(&policy->bydst, newpos); 1358 else 1359 hlist_add_head_rcu(&policy->bydst, chain); 1360 } 1361 1362 out_unlock: 1363 __xfrm_policy_inexact_flush(net); 1364 write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation); 1365 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1366 1367 mutex_unlock(&hash_resize_mutex); 1368 } 1369 1370 void xfrm_policy_hash_rebuild(struct net *net) 1371 { 1372 schedule_work(&net->xfrm.policy_hthresh.work); 1373 } 1374 EXPORT_SYMBOL(xfrm_policy_hash_rebuild); 1375 1376 /* Generate new index... KAME seems to generate them ordered by cost 1377 * of an absolute inpredictability of ordering of rules. This will not pass. */ 1378 static u32 xfrm_gen_index(struct net *net, int dir, u32 index) 1379 { 1380 for (;;) { 1381 struct hlist_head *list; 1382 struct xfrm_policy *p; 1383 u32 idx; 1384 int found; 1385 1386 if (!index) { 1387 idx = (net->xfrm.idx_generator | dir); 1388 net->xfrm.idx_generator += 8; 1389 } else { 1390 idx = index; 1391 index = 0; 1392 } 1393 1394 if (idx == 0) 1395 idx = 8; 1396 list = net->xfrm.policy_byidx + idx_hash(net, idx); 1397 found = 0; 1398 hlist_for_each_entry(p, list, byidx) { 1399 if (p->index == idx) { 1400 found = 1; 1401 break; 1402 } 1403 } 1404 if (!found) 1405 return idx; 1406 } 1407 } 1408 1409 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2) 1410 { 1411 u32 *p1 = (u32 *) s1; 1412 u32 *p2 = (u32 *) s2; 1413 int len = sizeof(struct xfrm_selector) / sizeof(u32); 1414 int i; 1415 1416 for (i = 0; i < len; i++) { 1417 if (p1[i] != p2[i]) 1418 return 1; 1419 } 1420 1421 return 0; 1422 } 1423 1424 static void xfrm_policy_requeue(struct xfrm_policy *old, 1425 struct xfrm_policy *new) 1426 { 1427 struct xfrm_policy_queue *pq = &old->polq; 1428 struct sk_buff_head list; 1429 1430 if (skb_queue_empty(&pq->hold_queue)) 1431 return; 1432 1433 __skb_queue_head_init(&list); 1434 1435 spin_lock_bh(&pq->hold_queue.lock); 1436 skb_queue_splice_init(&pq->hold_queue, &list); 1437 if (del_timer(&pq->hold_timer)) 1438 xfrm_pol_put(old); 1439 spin_unlock_bh(&pq->hold_queue.lock); 1440 1441 pq = &new->polq; 1442 1443 spin_lock_bh(&pq->hold_queue.lock); 1444 skb_queue_splice(&list, &pq->hold_queue); 1445 pq->timeout = XFRM_QUEUE_TMO_MIN; 1446 if (!mod_timer(&pq->hold_timer, jiffies)) 1447 xfrm_pol_hold(new); 1448 spin_unlock_bh(&pq->hold_queue.lock); 1449 } 1450 1451 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark, 1452 struct xfrm_policy *pol) 1453 { 1454 return mark->v == pol->mark.v && mark->m == pol->mark.m; 1455 } 1456 1457 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed) 1458 { 1459 const struct xfrm_pol_inexact_key *k = data; 1460 u32 a = k->type << 24 | k->dir << 16 | k->family; 1461 1462 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)), 1463 seed); 1464 } 1465 1466 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed) 1467 { 1468 const struct xfrm_pol_inexact_bin *b = data; 1469 1470 return xfrm_pol_bin_key(&b->k, 0, seed); 1471 } 1472 1473 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg, 1474 const void *ptr) 1475 { 1476 const struct xfrm_pol_inexact_key *key = arg->key; 1477 const struct xfrm_pol_inexact_bin *b = ptr; 1478 int ret; 1479 1480 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net))) 1481 return -1; 1482 1483 ret = b->k.dir ^ key->dir; 1484 if (ret) 1485 return ret; 1486 1487 ret = b->k.type ^ key->type; 1488 if (ret) 1489 return ret; 1490 1491 ret = b->k.family ^ key->family; 1492 if (ret) 1493 return ret; 1494 1495 return b->k.if_id ^ key->if_id; 1496 } 1497 1498 static const struct rhashtable_params xfrm_pol_inexact_params = { 1499 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head), 1500 .hashfn = xfrm_pol_bin_key, 1501 .obj_hashfn = xfrm_pol_bin_obj, 1502 .obj_cmpfn = xfrm_pol_bin_cmp, 1503 .automatic_shrinking = true, 1504 }; 1505 1506 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain, 1507 struct xfrm_policy *policy) 1508 { 1509 struct xfrm_policy *pol, *delpol = NULL; 1510 struct hlist_node *newpos = NULL; 1511 int i = 0; 1512 1513 hlist_for_each_entry(pol, chain, bydst_inexact_list) { 1514 if (pol->type == policy->type && 1515 pol->if_id == policy->if_id && 1516 !selector_cmp(&pol->selector, &policy->selector) && 1517 xfrm_policy_mark_match(&policy->mark, pol) && 1518 xfrm_sec_ctx_match(pol->security, policy->security) && 1519 !WARN_ON(delpol)) { 1520 delpol = pol; 1521 if (policy->priority > pol->priority) 1522 continue; 1523 } else if (policy->priority >= pol->priority) { 1524 newpos = &pol->bydst_inexact_list; 1525 continue; 1526 } 1527 if (delpol) 1528 break; 1529 } 1530 1531 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET) 1532 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos); 1533 else 1534 hlist_add_head_rcu(&policy->bydst_inexact_list, chain); 1535 1536 hlist_for_each_entry(pol, chain, bydst_inexact_list) { 1537 pol->pos = i; 1538 i++; 1539 } 1540 } 1541 1542 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain, 1543 struct xfrm_policy *policy, 1544 bool excl) 1545 { 1546 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL; 1547 1548 hlist_for_each_entry(pol, chain, bydst) { 1549 if (pol->type == policy->type && 1550 pol->if_id == policy->if_id && 1551 !selector_cmp(&pol->selector, &policy->selector) && 1552 xfrm_policy_mark_match(&policy->mark, pol) && 1553 xfrm_sec_ctx_match(pol->security, policy->security) && 1554 !WARN_ON(delpol)) { 1555 if (excl) 1556 return ERR_PTR(-EEXIST); 1557 delpol = pol; 1558 if (policy->priority > pol->priority) 1559 continue; 1560 } else if (policy->priority >= pol->priority) { 1561 newpos = pol; 1562 continue; 1563 } 1564 if (delpol) 1565 break; 1566 } 1567 1568 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET) 1569 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst); 1570 else 1571 /* Packet offload policies enter to the head 1572 * to speed-up lookups. 1573 */ 1574 hlist_add_head_rcu(&policy->bydst, chain); 1575 1576 return delpol; 1577 } 1578 1579 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl) 1580 { 1581 struct net *net = xp_net(policy); 1582 struct xfrm_policy *delpol; 1583 struct hlist_head *chain; 1584 1585 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1586 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir); 1587 if (chain) 1588 delpol = xfrm_policy_insert_list(chain, policy, excl); 1589 else 1590 delpol = xfrm_policy_inexact_insert(policy, dir, excl); 1591 1592 if (IS_ERR(delpol)) { 1593 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1594 return PTR_ERR(delpol); 1595 } 1596 1597 __xfrm_policy_link(policy, dir); 1598 1599 /* After previous checking, family can either be AF_INET or AF_INET6 */ 1600 if (policy->family == AF_INET) 1601 rt_genid_bump_ipv4(net); 1602 else 1603 rt_genid_bump_ipv6(net); 1604 1605 if (delpol) { 1606 xfrm_policy_requeue(delpol, policy); 1607 __xfrm_policy_unlink(delpol, dir); 1608 } 1609 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index); 1610 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index)); 1611 policy->curlft.add_time = ktime_get_real_seconds(); 1612 policy->curlft.use_time = 0; 1613 if (!mod_timer(&policy->timer, jiffies + HZ)) 1614 xfrm_pol_hold(policy); 1615 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1616 1617 if (delpol) 1618 xfrm_policy_kill(delpol); 1619 else if (xfrm_bydst_should_resize(net, dir, NULL)) 1620 schedule_work(&net->xfrm.policy_hash_work); 1621 1622 return 0; 1623 } 1624 EXPORT_SYMBOL(xfrm_policy_insert); 1625 1626 static struct xfrm_policy * 1627 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark, 1628 u32 if_id, u8 type, int dir, struct xfrm_selector *sel, 1629 struct xfrm_sec_ctx *ctx) 1630 { 1631 struct xfrm_policy *pol; 1632 1633 if (!chain) 1634 return NULL; 1635 1636 hlist_for_each_entry(pol, chain, bydst) { 1637 if (pol->type == type && 1638 pol->if_id == if_id && 1639 xfrm_policy_mark_match(mark, pol) && 1640 !selector_cmp(sel, &pol->selector) && 1641 xfrm_sec_ctx_match(ctx, pol->security)) 1642 return pol; 1643 } 1644 1645 return NULL; 1646 } 1647 1648 struct xfrm_policy * 1649 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id, 1650 u8 type, int dir, struct xfrm_selector *sel, 1651 struct xfrm_sec_ctx *ctx, int delete, int *err) 1652 { 1653 struct xfrm_pol_inexact_bin *bin = NULL; 1654 struct xfrm_policy *pol, *ret = NULL; 1655 struct hlist_head *chain; 1656 1657 *err = 0; 1658 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1659 chain = policy_hash_bysel(net, sel, sel->family, dir); 1660 if (!chain) { 1661 struct xfrm_pol_inexact_candidates cand; 1662 int i; 1663 1664 bin = xfrm_policy_inexact_lookup(net, type, 1665 sel->family, dir, if_id); 1666 if (!bin) { 1667 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1668 return NULL; 1669 } 1670 1671 if (!xfrm_policy_find_inexact_candidates(&cand, bin, 1672 &sel->saddr, 1673 &sel->daddr)) { 1674 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1675 return NULL; 1676 } 1677 1678 pol = NULL; 1679 for (i = 0; i < ARRAY_SIZE(cand.res); i++) { 1680 struct xfrm_policy *tmp; 1681 1682 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark, 1683 if_id, type, dir, 1684 sel, ctx); 1685 if (!tmp) 1686 continue; 1687 1688 if (!pol || tmp->pos < pol->pos) 1689 pol = tmp; 1690 } 1691 } else { 1692 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir, 1693 sel, ctx); 1694 } 1695 1696 if (pol) { 1697 xfrm_pol_hold(pol); 1698 if (delete) { 1699 *err = security_xfrm_policy_delete(pol->security); 1700 if (*err) { 1701 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1702 return pol; 1703 } 1704 __xfrm_policy_unlink(pol, dir); 1705 } 1706 ret = pol; 1707 } 1708 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1709 1710 if (ret && delete) 1711 xfrm_policy_kill(ret); 1712 if (bin && delete) 1713 xfrm_policy_inexact_prune_bin(bin); 1714 return ret; 1715 } 1716 EXPORT_SYMBOL(xfrm_policy_bysel_ctx); 1717 1718 struct xfrm_policy * 1719 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id, 1720 u8 type, int dir, u32 id, int delete, int *err) 1721 { 1722 struct xfrm_policy *pol, *ret; 1723 struct hlist_head *chain; 1724 1725 *err = -ENOENT; 1726 if (xfrm_policy_id2dir(id) != dir) 1727 return NULL; 1728 1729 *err = 0; 1730 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1731 chain = net->xfrm.policy_byidx + idx_hash(net, id); 1732 ret = NULL; 1733 hlist_for_each_entry(pol, chain, byidx) { 1734 if (pol->type == type && pol->index == id && 1735 pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) { 1736 xfrm_pol_hold(pol); 1737 if (delete) { 1738 *err = security_xfrm_policy_delete( 1739 pol->security); 1740 if (*err) { 1741 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1742 return pol; 1743 } 1744 __xfrm_policy_unlink(pol, dir); 1745 } 1746 ret = pol; 1747 break; 1748 } 1749 } 1750 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1751 1752 if (ret && delete) 1753 xfrm_policy_kill(ret); 1754 return ret; 1755 } 1756 EXPORT_SYMBOL(xfrm_policy_byid); 1757 1758 #ifdef CONFIG_SECURITY_NETWORK_XFRM 1759 static inline int 1760 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid) 1761 { 1762 struct xfrm_policy *pol; 1763 int err = 0; 1764 1765 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1766 if (pol->walk.dead || 1767 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX || 1768 pol->type != type) 1769 continue; 1770 1771 err = security_xfrm_policy_delete(pol->security); 1772 if (err) { 1773 xfrm_audit_policy_delete(pol, 0, task_valid); 1774 return err; 1775 } 1776 } 1777 return err; 1778 } 1779 1780 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net, 1781 struct net_device *dev, 1782 bool task_valid) 1783 { 1784 struct xfrm_policy *pol; 1785 int err = 0; 1786 1787 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1788 if (pol->walk.dead || 1789 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX || 1790 pol->xdo.dev != dev) 1791 continue; 1792 1793 err = security_xfrm_policy_delete(pol->security); 1794 if (err) { 1795 xfrm_audit_policy_delete(pol, 0, task_valid); 1796 return err; 1797 } 1798 } 1799 return err; 1800 } 1801 #else 1802 static inline int 1803 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid) 1804 { 1805 return 0; 1806 } 1807 1808 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net, 1809 struct net_device *dev, 1810 bool task_valid) 1811 { 1812 return 0; 1813 } 1814 #endif 1815 1816 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid) 1817 { 1818 int dir, err = 0, cnt = 0; 1819 struct xfrm_policy *pol; 1820 1821 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1822 1823 err = xfrm_policy_flush_secctx_check(net, type, task_valid); 1824 if (err) 1825 goto out; 1826 1827 again: 1828 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1829 if (pol->walk.dead) 1830 continue; 1831 1832 dir = xfrm_policy_id2dir(pol->index); 1833 if (dir >= XFRM_POLICY_MAX || 1834 pol->type != type) 1835 continue; 1836 1837 __xfrm_policy_unlink(pol, dir); 1838 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1839 cnt++; 1840 xfrm_audit_policy_delete(pol, 1, task_valid); 1841 xfrm_policy_kill(pol); 1842 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1843 goto again; 1844 } 1845 if (cnt) 1846 __xfrm_policy_inexact_flush(net); 1847 else 1848 err = -ESRCH; 1849 out: 1850 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1851 return err; 1852 } 1853 EXPORT_SYMBOL(xfrm_policy_flush); 1854 1855 int xfrm_dev_policy_flush(struct net *net, struct net_device *dev, 1856 bool task_valid) 1857 { 1858 int dir, err = 0, cnt = 0; 1859 struct xfrm_policy *pol; 1860 1861 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1862 1863 err = xfrm_dev_policy_flush_secctx_check(net, dev, task_valid); 1864 if (err) 1865 goto out; 1866 1867 again: 1868 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1869 if (pol->walk.dead) 1870 continue; 1871 1872 dir = xfrm_policy_id2dir(pol->index); 1873 if (dir >= XFRM_POLICY_MAX || 1874 pol->xdo.dev != dev) 1875 continue; 1876 1877 __xfrm_policy_unlink(pol, dir); 1878 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1879 cnt++; 1880 xfrm_audit_policy_delete(pol, 1, task_valid); 1881 xfrm_policy_kill(pol); 1882 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1883 goto again; 1884 } 1885 if (cnt) 1886 __xfrm_policy_inexact_flush(net); 1887 else 1888 err = -ESRCH; 1889 out: 1890 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1891 return err; 1892 } 1893 EXPORT_SYMBOL(xfrm_dev_policy_flush); 1894 1895 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk, 1896 int (*func)(struct xfrm_policy *, int, int, void*), 1897 void *data) 1898 { 1899 struct xfrm_policy *pol; 1900 struct xfrm_policy_walk_entry *x; 1901 int error = 0; 1902 1903 if (walk->type >= XFRM_POLICY_TYPE_MAX && 1904 walk->type != XFRM_POLICY_TYPE_ANY) 1905 return -EINVAL; 1906 1907 if (list_empty(&walk->walk.all) && walk->seq != 0) 1908 return 0; 1909 1910 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1911 if (list_empty(&walk->walk.all)) 1912 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all); 1913 else 1914 x = list_first_entry(&walk->walk.all, 1915 struct xfrm_policy_walk_entry, all); 1916 1917 list_for_each_entry_from(x, &net->xfrm.policy_all, all) { 1918 if (x->dead) 1919 continue; 1920 pol = container_of(x, struct xfrm_policy, walk); 1921 if (walk->type != XFRM_POLICY_TYPE_ANY && 1922 walk->type != pol->type) 1923 continue; 1924 error = func(pol, xfrm_policy_id2dir(pol->index), 1925 walk->seq, data); 1926 if (error) { 1927 list_move_tail(&walk->walk.all, &x->all); 1928 goto out; 1929 } 1930 walk->seq++; 1931 } 1932 if (walk->seq == 0) { 1933 error = -ENOENT; 1934 goto out; 1935 } 1936 list_del_init(&walk->walk.all); 1937 out: 1938 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1939 return error; 1940 } 1941 EXPORT_SYMBOL(xfrm_policy_walk); 1942 1943 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type) 1944 { 1945 INIT_LIST_HEAD(&walk->walk.all); 1946 walk->walk.dead = 1; 1947 walk->type = type; 1948 walk->seq = 0; 1949 } 1950 EXPORT_SYMBOL(xfrm_policy_walk_init); 1951 1952 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net) 1953 { 1954 if (list_empty(&walk->walk.all)) 1955 return; 1956 1957 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */ 1958 list_del(&walk->walk.all); 1959 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1960 } 1961 EXPORT_SYMBOL(xfrm_policy_walk_done); 1962 1963 /* 1964 * Find policy to apply to this flow. 1965 * 1966 * Returns 0 if policy found, else an -errno. 1967 */ 1968 static int xfrm_policy_match(const struct xfrm_policy *pol, 1969 const struct flowi *fl, 1970 u8 type, u16 family, u32 if_id) 1971 { 1972 const struct xfrm_selector *sel = &pol->selector; 1973 int ret = -ESRCH; 1974 bool match; 1975 1976 if (pol->family != family || 1977 pol->if_id != if_id || 1978 (fl->flowi_mark & pol->mark.m) != pol->mark.v || 1979 pol->type != type) 1980 return ret; 1981 1982 match = xfrm_selector_match(sel, fl, family); 1983 if (match) 1984 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid); 1985 return ret; 1986 } 1987 1988 static struct xfrm_pol_inexact_node * 1989 xfrm_policy_lookup_inexact_addr(const struct rb_root *r, 1990 seqcount_spinlock_t *count, 1991 const xfrm_address_t *addr, u16 family) 1992 { 1993 const struct rb_node *parent; 1994 int seq; 1995 1996 again: 1997 seq = read_seqcount_begin(count); 1998 1999 parent = rcu_dereference_raw(r->rb_node); 2000 while (parent) { 2001 struct xfrm_pol_inexact_node *node; 2002 int delta; 2003 2004 node = rb_entry(parent, struct xfrm_pol_inexact_node, node); 2005 2006 delta = xfrm_policy_addr_delta(addr, &node->addr, 2007 node->prefixlen, family); 2008 if (delta < 0) { 2009 parent = rcu_dereference_raw(parent->rb_left); 2010 continue; 2011 } else if (delta > 0) { 2012 parent = rcu_dereference_raw(parent->rb_right); 2013 continue; 2014 } 2015 2016 return node; 2017 } 2018 2019 if (read_seqcount_retry(count, seq)) 2020 goto again; 2021 2022 return NULL; 2023 } 2024 2025 static bool 2026 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand, 2027 struct xfrm_pol_inexact_bin *b, 2028 const xfrm_address_t *saddr, 2029 const xfrm_address_t *daddr) 2030 { 2031 struct xfrm_pol_inexact_node *n; 2032 u16 family; 2033 2034 if (!b) 2035 return false; 2036 2037 family = b->k.family; 2038 memset(cand, 0, sizeof(*cand)); 2039 cand->res[XFRM_POL_CAND_ANY] = &b->hhead; 2040 2041 n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr, 2042 family); 2043 if (n) { 2044 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead; 2045 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr, 2046 family); 2047 if (n) 2048 cand->res[XFRM_POL_CAND_BOTH] = &n->hhead; 2049 } 2050 2051 n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr, 2052 family); 2053 if (n) 2054 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead; 2055 2056 return true; 2057 } 2058 2059 static struct xfrm_pol_inexact_bin * 2060 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family, 2061 u8 dir, u32 if_id) 2062 { 2063 struct xfrm_pol_inexact_key k = { 2064 .family = family, 2065 .type = type, 2066 .dir = dir, 2067 .if_id = if_id, 2068 }; 2069 2070 write_pnet(&k.net, net); 2071 2072 return rhashtable_lookup(&xfrm_policy_inexact_table, &k, 2073 xfrm_pol_inexact_params); 2074 } 2075 2076 static struct xfrm_pol_inexact_bin * 2077 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, 2078 u8 dir, u32 if_id) 2079 { 2080 struct xfrm_pol_inexact_bin *bin; 2081 2082 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 2083 2084 rcu_read_lock(); 2085 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id); 2086 rcu_read_unlock(); 2087 2088 return bin; 2089 } 2090 2091 static struct xfrm_policy * 2092 __xfrm_policy_eval_candidates(struct hlist_head *chain, 2093 struct xfrm_policy *prefer, 2094 const struct flowi *fl, 2095 u8 type, u16 family, u32 if_id) 2096 { 2097 u32 priority = prefer ? prefer->priority : ~0u; 2098 struct xfrm_policy *pol; 2099 2100 if (!chain) 2101 return NULL; 2102 2103 hlist_for_each_entry_rcu(pol, chain, bydst) { 2104 int err; 2105 2106 if (pol->priority > priority) 2107 break; 2108 2109 err = xfrm_policy_match(pol, fl, type, family, if_id); 2110 if (err) { 2111 if (err != -ESRCH) 2112 return ERR_PTR(err); 2113 2114 continue; 2115 } 2116 2117 if (prefer) { 2118 /* matches. Is it older than *prefer? */ 2119 if (pol->priority == priority && 2120 prefer->pos < pol->pos) 2121 return prefer; 2122 } 2123 2124 return pol; 2125 } 2126 2127 return NULL; 2128 } 2129 2130 static struct xfrm_policy * 2131 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand, 2132 struct xfrm_policy *prefer, 2133 const struct flowi *fl, 2134 u8 type, u16 family, u32 if_id) 2135 { 2136 struct xfrm_policy *tmp; 2137 int i; 2138 2139 for (i = 0; i < ARRAY_SIZE(cand->res); i++) { 2140 tmp = __xfrm_policy_eval_candidates(cand->res[i], 2141 prefer, 2142 fl, type, family, if_id); 2143 if (!tmp) 2144 continue; 2145 2146 if (IS_ERR(tmp)) 2147 return tmp; 2148 prefer = tmp; 2149 } 2150 2151 return prefer; 2152 } 2153 2154 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type, 2155 const struct flowi *fl, 2156 u16 family, u8 dir, 2157 u32 if_id) 2158 { 2159 struct xfrm_pol_inexact_candidates cand; 2160 const xfrm_address_t *daddr, *saddr; 2161 struct xfrm_pol_inexact_bin *bin; 2162 struct xfrm_policy *pol, *ret; 2163 struct hlist_head *chain; 2164 unsigned int sequence; 2165 int err; 2166 2167 daddr = xfrm_flowi_daddr(fl, family); 2168 saddr = xfrm_flowi_saddr(fl, family); 2169 if (unlikely(!daddr || !saddr)) 2170 return NULL; 2171 2172 rcu_read_lock(); 2173 retry: 2174 do { 2175 sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation); 2176 chain = policy_hash_direct(net, daddr, saddr, family, dir); 2177 } while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence)); 2178 2179 ret = NULL; 2180 hlist_for_each_entry_rcu(pol, chain, bydst) { 2181 err = xfrm_policy_match(pol, fl, type, family, if_id); 2182 if (err) { 2183 if (err == -ESRCH) 2184 continue; 2185 else { 2186 ret = ERR_PTR(err); 2187 goto fail; 2188 } 2189 } else { 2190 ret = pol; 2191 break; 2192 } 2193 } 2194 if (ret && ret->xdo.type == XFRM_DEV_OFFLOAD_PACKET) 2195 goto skip_inexact; 2196 2197 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id); 2198 if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr, 2199 daddr)) 2200 goto skip_inexact; 2201 2202 pol = xfrm_policy_eval_candidates(&cand, ret, fl, type, 2203 family, if_id); 2204 if (pol) { 2205 ret = pol; 2206 if (IS_ERR(pol)) 2207 goto fail; 2208 } 2209 2210 skip_inexact: 2211 if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence)) 2212 goto retry; 2213 2214 if (ret && !xfrm_pol_hold_rcu(ret)) 2215 goto retry; 2216 fail: 2217 rcu_read_unlock(); 2218 2219 return ret; 2220 } 2221 2222 static struct xfrm_policy *xfrm_policy_lookup(struct net *net, 2223 const struct flowi *fl, 2224 u16 family, u8 dir, u32 if_id) 2225 { 2226 #ifdef CONFIG_XFRM_SUB_POLICY 2227 struct xfrm_policy *pol; 2228 2229 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, 2230 dir, if_id); 2231 if (pol != NULL) 2232 return pol; 2233 #endif 2234 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, 2235 dir, if_id); 2236 } 2237 2238 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir, 2239 const struct flowi *fl, 2240 u16 family, u32 if_id) 2241 { 2242 struct xfrm_policy *pol; 2243 2244 rcu_read_lock(); 2245 again: 2246 pol = rcu_dereference(sk->sk_policy[dir]); 2247 if (pol != NULL) { 2248 bool match; 2249 int err = 0; 2250 2251 if (pol->family != family) { 2252 pol = NULL; 2253 goto out; 2254 } 2255 2256 match = xfrm_selector_match(&pol->selector, fl, family); 2257 if (match) { 2258 if ((READ_ONCE(sk->sk_mark) & pol->mark.m) != pol->mark.v || 2259 pol->if_id != if_id) { 2260 pol = NULL; 2261 goto out; 2262 } 2263 err = security_xfrm_policy_lookup(pol->security, 2264 fl->flowi_secid); 2265 if (!err) { 2266 if (!xfrm_pol_hold_rcu(pol)) 2267 goto again; 2268 } else if (err == -ESRCH) { 2269 pol = NULL; 2270 } else { 2271 pol = ERR_PTR(err); 2272 } 2273 } else 2274 pol = NULL; 2275 } 2276 out: 2277 rcu_read_unlock(); 2278 return pol; 2279 } 2280 2281 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir) 2282 { 2283 struct net *net = xp_net(pol); 2284 2285 list_add(&pol->walk.all, &net->xfrm.policy_all); 2286 net->xfrm.policy_count[dir]++; 2287 xfrm_pol_hold(pol); 2288 } 2289 2290 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol, 2291 int dir) 2292 { 2293 struct net *net = xp_net(pol); 2294 2295 if (list_empty(&pol->walk.all)) 2296 return NULL; 2297 2298 /* Socket policies are not hashed. */ 2299 if (!hlist_unhashed(&pol->bydst)) { 2300 hlist_del_rcu(&pol->bydst); 2301 hlist_del_init(&pol->bydst_inexact_list); 2302 hlist_del(&pol->byidx); 2303 } 2304 2305 list_del_init(&pol->walk.all); 2306 net->xfrm.policy_count[dir]--; 2307 2308 return pol; 2309 } 2310 2311 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir) 2312 { 2313 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir); 2314 } 2315 2316 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir) 2317 { 2318 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir); 2319 } 2320 2321 int xfrm_policy_delete(struct xfrm_policy *pol, int dir) 2322 { 2323 struct net *net = xp_net(pol); 2324 2325 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 2326 pol = __xfrm_policy_unlink(pol, dir); 2327 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 2328 if (pol) { 2329 xfrm_policy_kill(pol); 2330 return 0; 2331 } 2332 return -ENOENT; 2333 } 2334 EXPORT_SYMBOL(xfrm_policy_delete); 2335 2336 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol) 2337 { 2338 struct net *net = sock_net(sk); 2339 struct xfrm_policy *old_pol; 2340 2341 #ifdef CONFIG_XFRM_SUB_POLICY 2342 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN) 2343 return -EINVAL; 2344 #endif 2345 2346 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 2347 old_pol = rcu_dereference_protected(sk->sk_policy[dir], 2348 lockdep_is_held(&net->xfrm.xfrm_policy_lock)); 2349 if (pol) { 2350 pol->curlft.add_time = ktime_get_real_seconds(); 2351 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0); 2352 xfrm_sk_policy_link(pol, dir); 2353 } 2354 rcu_assign_pointer(sk->sk_policy[dir], pol); 2355 if (old_pol) { 2356 if (pol) 2357 xfrm_policy_requeue(old_pol, pol); 2358 2359 /* Unlinking succeeds always. This is the only function 2360 * allowed to delete or replace socket policy. 2361 */ 2362 xfrm_sk_policy_unlink(old_pol, dir); 2363 } 2364 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 2365 2366 if (old_pol) { 2367 xfrm_policy_kill(old_pol); 2368 } 2369 return 0; 2370 } 2371 2372 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir) 2373 { 2374 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC); 2375 struct net *net = xp_net(old); 2376 2377 if (newp) { 2378 newp->selector = old->selector; 2379 if (security_xfrm_policy_clone(old->security, 2380 &newp->security)) { 2381 kfree(newp); 2382 return NULL; /* ENOMEM */ 2383 } 2384 newp->lft = old->lft; 2385 newp->curlft = old->curlft; 2386 newp->mark = old->mark; 2387 newp->if_id = old->if_id; 2388 newp->action = old->action; 2389 newp->flags = old->flags; 2390 newp->xfrm_nr = old->xfrm_nr; 2391 newp->index = old->index; 2392 newp->type = old->type; 2393 newp->family = old->family; 2394 memcpy(newp->xfrm_vec, old->xfrm_vec, 2395 newp->xfrm_nr*sizeof(struct xfrm_tmpl)); 2396 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 2397 xfrm_sk_policy_link(newp, dir); 2398 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 2399 xfrm_pol_put(newp); 2400 } 2401 return newp; 2402 } 2403 2404 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk) 2405 { 2406 const struct xfrm_policy *p; 2407 struct xfrm_policy *np; 2408 int i, ret = 0; 2409 2410 rcu_read_lock(); 2411 for (i = 0; i < 2; i++) { 2412 p = rcu_dereference(osk->sk_policy[i]); 2413 if (p) { 2414 np = clone_policy(p, i); 2415 if (unlikely(!np)) { 2416 ret = -ENOMEM; 2417 break; 2418 } 2419 rcu_assign_pointer(sk->sk_policy[i], np); 2420 } 2421 } 2422 rcu_read_unlock(); 2423 return ret; 2424 } 2425 2426 static int 2427 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local, 2428 xfrm_address_t *remote, unsigned short family, u32 mark) 2429 { 2430 int err; 2431 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 2432 2433 if (unlikely(afinfo == NULL)) 2434 return -EINVAL; 2435 err = afinfo->get_saddr(net, oif, local, remote, mark); 2436 rcu_read_unlock(); 2437 return err; 2438 } 2439 2440 /* Resolve list of templates for the flow, given policy. */ 2441 2442 static int 2443 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl, 2444 struct xfrm_state **xfrm, unsigned short family) 2445 { 2446 struct net *net = xp_net(policy); 2447 int nx; 2448 int i, error; 2449 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family); 2450 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family); 2451 xfrm_address_t tmp; 2452 2453 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) { 2454 struct xfrm_state *x; 2455 xfrm_address_t *remote = daddr; 2456 xfrm_address_t *local = saddr; 2457 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i]; 2458 2459 if (tmpl->mode == XFRM_MODE_TUNNEL || 2460 tmpl->mode == XFRM_MODE_BEET) { 2461 remote = &tmpl->id.daddr; 2462 local = &tmpl->saddr; 2463 if (xfrm_addr_any(local, tmpl->encap_family)) { 2464 error = xfrm_get_saddr(net, fl->flowi_oif, 2465 &tmp, remote, 2466 tmpl->encap_family, 0); 2467 if (error) 2468 goto fail; 2469 local = &tmp; 2470 } 2471 } 2472 2473 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, 2474 family, policy->if_id); 2475 2476 if (x && x->km.state == XFRM_STATE_VALID) { 2477 xfrm[nx++] = x; 2478 daddr = remote; 2479 saddr = local; 2480 continue; 2481 } 2482 if (x) { 2483 error = (x->km.state == XFRM_STATE_ERROR ? 2484 -EINVAL : -EAGAIN); 2485 xfrm_state_put(x); 2486 } else if (error == -ESRCH) { 2487 error = -EAGAIN; 2488 } 2489 2490 if (!tmpl->optional) 2491 goto fail; 2492 } 2493 return nx; 2494 2495 fail: 2496 for (nx--; nx >= 0; nx--) 2497 xfrm_state_put(xfrm[nx]); 2498 return error; 2499 } 2500 2501 static int 2502 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl, 2503 struct xfrm_state **xfrm, unsigned short family) 2504 { 2505 struct xfrm_state *tp[XFRM_MAX_DEPTH]; 2506 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm; 2507 int cnx = 0; 2508 int error; 2509 int ret; 2510 int i; 2511 2512 for (i = 0; i < npols; i++) { 2513 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) { 2514 error = -ENOBUFS; 2515 goto fail; 2516 } 2517 2518 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family); 2519 if (ret < 0) { 2520 error = ret; 2521 goto fail; 2522 } else 2523 cnx += ret; 2524 } 2525 2526 /* found states are sorted for outbound processing */ 2527 if (npols > 1) 2528 xfrm_state_sort(xfrm, tpp, cnx, family); 2529 2530 return cnx; 2531 2532 fail: 2533 for (cnx--; cnx >= 0; cnx--) 2534 xfrm_state_put(tpp[cnx]); 2535 return error; 2536 2537 } 2538 2539 static int xfrm_get_tos(const struct flowi *fl, int family) 2540 { 2541 if (family == AF_INET) 2542 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos; 2543 2544 return 0; 2545 } 2546 2547 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family) 2548 { 2549 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 2550 struct dst_ops *dst_ops; 2551 struct xfrm_dst *xdst; 2552 2553 if (!afinfo) 2554 return ERR_PTR(-EINVAL); 2555 2556 switch (family) { 2557 case AF_INET: 2558 dst_ops = &net->xfrm.xfrm4_dst_ops; 2559 break; 2560 #if IS_ENABLED(CONFIG_IPV6) 2561 case AF_INET6: 2562 dst_ops = &net->xfrm.xfrm6_dst_ops; 2563 break; 2564 #endif 2565 default: 2566 BUG(); 2567 } 2568 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0); 2569 2570 if (likely(xdst)) { 2571 memset_after(xdst, 0, u.dst); 2572 } else 2573 xdst = ERR_PTR(-ENOBUFS); 2574 2575 rcu_read_unlock(); 2576 2577 return xdst; 2578 } 2579 2580 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst, 2581 int nfheader_len) 2582 { 2583 if (dst->ops->family == AF_INET6) { 2584 struct rt6_info *rt = (struct rt6_info *)dst; 2585 path->path_cookie = rt6_get_cookie(rt); 2586 path->u.rt6.rt6i_nfheader_len = nfheader_len; 2587 } 2588 } 2589 2590 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev, 2591 const struct flowi *fl) 2592 { 2593 const struct xfrm_policy_afinfo *afinfo = 2594 xfrm_policy_get_afinfo(xdst->u.dst.ops->family); 2595 int err; 2596 2597 if (!afinfo) 2598 return -EINVAL; 2599 2600 err = afinfo->fill_dst(xdst, dev, fl); 2601 2602 rcu_read_unlock(); 2603 2604 return err; 2605 } 2606 2607 2608 /* Allocate chain of dst_entry's, attach known xfrm's, calculate 2609 * all the metrics... Shortly, bundle a bundle. 2610 */ 2611 2612 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy, 2613 struct xfrm_state **xfrm, 2614 struct xfrm_dst **bundle, 2615 int nx, 2616 const struct flowi *fl, 2617 struct dst_entry *dst) 2618 { 2619 const struct xfrm_state_afinfo *afinfo; 2620 const struct xfrm_mode *inner_mode; 2621 struct net *net = xp_net(policy); 2622 unsigned long now = jiffies; 2623 struct net_device *dev; 2624 struct xfrm_dst *xdst_prev = NULL; 2625 struct xfrm_dst *xdst0 = NULL; 2626 int i = 0; 2627 int err; 2628 int header_len = 0; 2629 int nfheader_len = 0; 2630 int trailer_len = 0; 2631 int tos; 2632 int family = policy->selector.family; 2633 xfrm_address_t saddr, daddr; 2634 2635 xfrm_flowi_addr_get(fl, &saddr, &daddr, family); 2636 2637 tos = xfrm_get_tos(fl, family); 2638 2639 dst_hold(dst); 2640 2641 for (; i < nx; i++) { 2642 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family); 2643 struct dst_entry *dst1 = &xdst->u.dst; 2644 2645 err = PTR_ERR(xdst); 2646 if (IS_ERR(xdst)) { 2647 dst_release(dst); 2648 goto put_states; 2649 } 2650 2651 bundle[i] = xdst; 2652 if (!xdst_prev) 2653 xdst0 = xdst; 2654 else 2655 /* Ref count is taken during xfrm_alloc_dst() 2656 * No need to do dst_clone() on dst1 2657 */ 2658 xfrm_dst_set_child(xdst_prev, &xdst->u.dst); 2659 2660 if (xfrm[i]->sel.family == AF_UNSPEC) { 2661 inner_mode = xfrm_ip2inner_mode(xfrm[i], 2662 xfrm_af2proto(family)); 2663 if (!inner_mode) { 2664 err = -EAFNOSUPPORT; 2665 dst_release(dst); 2666 goto put_states; 2667 } 2668 } else 2669 inner_mode = &xfrm[i]->inner_mode; 2670 2671 xdst->route = dst; 2672 dst_copy_metrics(dst1, dst); 2673 2674 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) { 2675 __u32 mark = 0; 2676 int oif; 2677 2678 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m) 2679 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]); 2680 2681 if (xfrm[i]->xso.type != XFRM_DEV_OFFLOAD_PACKET) 2682 family = xfrm[i]->props.family; 2683 2684 oif = fl->flowi_oif ? : fl->flowi_l3mdev; 2685 dst = xfrm_dst_lookup(xfrm[i], tos, oif, 2686 &saddr, &daddr, family, mark); 2687 err = PTR_ERR(dst); 2688 if (IS_ERR(dst)) 2689 goto put_states; 2690 } else 2691 dst_hold(dst); 2692 2693 dst1->xfrm = xfrm[i]; 2694 xdst->xfrm_genid = xfrm[i]->genid; 2695 2696 dst1->obsolete = DST_OBSOLETE_FORCE_CHK; 2697 dst1->lastuse = now; 2698 2699 dst1->input = dst_discard; 2700 2701 rcu_read_lock(); 2702 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family); 2703 if (likely(afinfo)) 2704 dst1->output = afinfo->output; 2705 else 2706 dst1->output = dst_discard_out; 2707 rcu_read_unlock(); 2708 2709 xdst_prev = xdst; 2710 2711 header_len += xfrm[i]->props.header_len; 2712 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT) 2713 nfheader_len += xfrm[i]->props.header_len; 2714 trailer_len += xfrm[i]->props.trailer_len; 2715 } 2716 2717 xfrm_dst_set_child(xdst_prev, dst); 2718 xdst0->path = dst; 2719 2720 err = -ENODEV; 2721 dev = dst->dev; 2722 if (!dev) 2723 goto free_dst; 2724 2725 xfrm_init_path(xdst0, dst, nfheader_len); 2726 xfrm_init_pmtu(bundle, nx); 2727 2728 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst; 2729 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) { 2730 err = xfrm_fill_dst(xdst_prev, dev, fl); 2731 if (err) 2732 goto free_dst; 2733 2734 xdst_prev->u.dst.header_len = header_len; 2735 xdst_prev->u.dst.trailer_len = trailer_len; 2736 header_len -= xdst_prev->u.dst.xfrm->props.header_len; 2737 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len; 2738 } 2739 2740 return &xdst0->u.dst; 2741 2742 put_states: 2743 for (; i < nx; i++) 2744 xfrm_state_put(xfrm[i]); 2745 free_dst: 2746 if (xdst0) 2747 dst_release_immediate(&xdst0->u.dst); 2748 2749 return ERR_PTR(err); 2750 } 2751 2752 static int xfrm_expand_policies(const struct flowi *fl, u16 family, 2753 struct xfrm_policy **pols, 2754 int *num_pols, int *num_xfrms) 2755 { 2756 int i; 2757 2758 if (*num_pols == 0 || !pols[0]) { 2759 *num_pols = 0; 2760 *num_xfrms = 0; 2761 return 0; 2762 } 2763 if (IS_ERR(pols[0])) { 2764 *num_pols = 0; 2765 return PTR_ERR(pols[0]); 2766 } 2767 2768 *num_xfrms = pols[0]->xfrm_nr; 2769 2770 #ifdef CONFIG_XFRM_SUB_POLICY 2771 if (pols[0]->action == XFRM_POLICY_ALLOW && 2772 pols[0]->type != XFRM_POLICY_TYPE_MAIN) { 2773 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]), 2774 XFRM_POLICY_TYPE_MAIN, 2775 fl, family, 2776 XFRM_POLICY_OUT, 2777 pols[0]->if_id); 2778 if (pols[1]) { 2779 if (IS_ERR(pols[1])) { 2780 xfrm_pols_put(pols, *num_pols); 2781 *num_pols = 0; 2782 return PTR_ERR(pols[1]); 2783 } 2784 (*num_pols)++; 2785 (*num_xfrms) += pols[1]->xfrm_nr; 2786 } 2787 } 2788 #endif 2789 for (i = 0; i < *num_pols; i++) { 2790 if (pols[i]->action != XFRM_POLICY_ALLOW) { 2791 *num_xfrms = -1; 2792 break; 2793 } 2794 } 2795 2796 return 0; 2797 2798 } 2799 2800 static struct xfrm_dst * 2801 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols, 2802 const struct flowi *fl, u16 family, 2803 struct dst_entry *dst_orig) 2804 { 2805 struct net *net = xp_net(pols[0]); 2806 struct xfrm_state *xfrm[XFRM_MAX_DEPTH]; 2807 struct xfrm_dst *bundle[XFRM_MAX_DEPTH]; 2808 struct xfrm_dst *xdst; 2809 struct dst_entry *dst; 2810 int err; 2811 2812 /* Try to instantiate a bundle */ 2813 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family); 2814 if (err <= 0) { 2815 if (err == 0) 2816 return NULL; 2817 2818 if (err != -EAGAIN) 2819 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR); 2820 return ERR_PTR(err); 2821 } 2822 2823 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig); 2824 if (IS_ERR(dst)) { 2825 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR); 2826 return ERR_CAST(dst); 2827 } 2828 2829 xdst = (struct xfrm_dst *)dst; 2830 xdst->num_xfrms = err; 2831 xdst->num_pols = num_pols; 2832 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols); 2833 xdst->policy_genid = atomic_read(&pols[0]->genid); 2834 2835 return xdst; 2836 } 2837 2838 static void xfrm_policy_queue_process(struct timer_list *t) 2839 { 2840 struct sk_buff *skb; 2841 struct sock *sk; 2842 struct dst_entry *dst; 2843 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer); 2844 struct net *net = xp_net(pol); 2845 struct xfrm_policy_queue *pq = &pol->polq; 2846 struct flowi fl; 2847 struct sk_buff_head list; 2848 __u32 skb_mark; 2849 2850 spin_lock(&pq->hold_queue.lock); 2851 skb = skb_peek(&pq->hold_queue); 2852 if (!skb) { 2853 spin_unlock(&pq->hold_queue.lock); 2854 goto out; 2855 } 2856 dst = skb_dst(skb); 2857 sk = skb->sk; 2858 2859 /* Fixup the mark to support VTI. */ 2860 skb_mark = skb->mark; 2861 skb->mark = pol->mark.v; 2862 xfrm_decode_session(skb, &fl, dst->ops->family); 2863 skb->mark = skb_mark; 2864 spin_unlock(&pq->hold_queue.lock); 2865 2866 dst_hold(xfrm_dst_path(dst)); 2867 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE); 2868 if (IS_ERR(dst)) 2869 goto purge_queue; 2870 2871 if (dst->flags & DST_XFRM_QUEUE) { 2872 dst_release(dst); 2873 2874 if (pq->timeout >= XFRM_QUEUE_TMO_MAX) 2875 goto purge_queue; 2876 2877 pq->timeout = pq->timeout << 1; 2878 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout)) 2879 xfrm_pol_hold(pol); 2880 goto out; 2881 } 2882 2883 dst_release(dst); 2884 2885 __skb_queue_head_init(&list); 2886 2887 spin_lock(&pq->hold_queue.lock); 2888 pq->timeout = 0; 2889 skb_queue_splice_init(&pq->hold_queue, &list); 2890 spin_unlock(&pq->hold_queue.lock); 2891 2892 while (!skb_queue_empty(&list)) { 2893 skb = __skb_dequeue(&list); 2894 2895 /* Fixup the mark to support VTI. */ 2896 skb_mark = skb->mark; 2897 skb->mark = pol->mark.v; 2898 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family); 2899 skb->mark = skb_mark; 2900 2901 dst_hold(xfrm_dst_path(skb_dst(skb))); 2902 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0); 2903 if (IS_ERR(dst)) { 2904 kfree_skb(skb); 2905 continue; 2906 } 2907 2908 nf_reset_ct(skb); 2909 skb_dst_drop(skb); 2910 skb_dst_set(skb, dst); 2911 2912 dst_output(net, skb->sk, skb); 2913 } 2914 2915 out: 2916 xfrm_pol_put(pol); 2917 return; 2918 2919 purge_queue: 2920 pq->timeout = 0; 2921 skb_queue_purge(&pq->hold_queue); 2922 xfrm_pol_put(pol); 2923 } 2924 2925 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb) 2926 { 2927 unsigned long sched_next; 2928 struct dst_entry *dst = skb_dst(skb); 2929 struct xfrm_dst *xdst = (struct xfrm_dst *) dst; 2930 struct xfrm_policy *pol = xdst->pols[0]; 2931 struct xfrm_policy_queue *pq = &pol->polq; 2932 2933 if (unlikely(skb_fclone_busy(sk, skb))) { 2934 kfree_skb(skb); 2935 return 0; 2936 } 2937 2938 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) { 2939 kfree_skb(skb); 2940 return -EAGAIN; 2941 } 2942 2943 skb_dst_force(skb); 2944 2945 spin_lock_bh(&pq->hold_queue.lock); 2946 2947 if (!pq->timeout) 2948 pq->timeout = XFRM_QUEUE_TMO_MIN; 2949 2950 sched_next = jiffies + pq->timeout; 2951 2952 if (del_timer(&pq->hold_timer)) { 2953 if (time_before(pq->hold_timer.expires, sched_next)) 2954 sched_next = pq->hold_timer.expires; 2955 xfrm_pol_put(pol); 2956 } 2957 2958 __skb_queue_tail(&pq->hold_queue, skb); 2959 if (!mod_timer(&pq->hold_timer, sched_next)) 2960 xfrm_pol_hold(pol); 2961 2962 spin_unlock_bh(&pq->hold_queue.lock); 2963 2964 return 0; 2965 } 2966 2967 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net, 2968 struct xfrm_flo *xflo, 2969 const struct flowi *fl, 2970 int num_xfrms, 2971 u16 family) 2972 { 2973 int err; 2974 struct net_device *dev; 2975 struct dst_entry *dst; 2976 struct dst_entry *dst1; 2977 struct xfrm_dst *xdst; 2978 2979 xdst = xfrm_alloc_dst(net, family); 2980 if (IS_ERR(xdst)) 2981 return xdst; 2982 2983 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) || 2984 net->xfrm.sysctl_larval_drop || 2985 num_xfrms <= 0) 2986 return xdst; 2987 2988 dst = xflo->dst_orig; 2989 dst1 = &xdst->u.dst; 2990 dst_hold(dst); 2991 xdst->route = dst; 2992 2993 dst_copy_metrics(dst1, dst); 2994 2995 dst1->obsolete = DST_OBSOLETE_FORCE_CHK; 2996 dst1->flags |= DST_XFRM_QUEUE; 2997 dst1->lastuse = jiffies; 2998 2999 dst1->input = dst_discard; 3000 dst1->output = xdst_queue_output; 3001 3002 dst_hold(dst); 3003 xfrm_dst_set_child(xdst, dst); 3004 xdst->path = dst; 3005 3006 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0); 3007 3008 err = -ENODEV; 3009 dev = dst->dev; 3010 if (!dev) 3011 goto free_dst; 3012 3013 err = xfrm_fill_dst(xdst, dev, fl); 3014 if (err) 3015 goto free_dst; 3016 3017 out: 3018 return xdst; 3019 3020 free_dst: 3021 dst_release(dst1); 3022 xdst = ERR_PTR(err); 3023 goto out; 3024 } 3025 3026 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net, 3027 const struct flowi *fl, 3028 u16 family, u8 dir, 3029 struct xfrm_flo *xflo, u32 if_id) 3030 { 3031 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 3032 int num_pols = 0, num_xfrms = 0, err; 3033 struct xfrm_dst *xdst; 3034 3035 /* Resolve policies to use if we couldn't get them from 3036 * previous cache entry */ 3037 num_pols = 1; 3038 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id); 3039 err = xfrm_expand_policies(fl, family, pols, 3040 &num_pols, &num_xfrms); 3041 if (err < 0) 3042 goto inc_error; 3043 if (num_pols == 0) 3044 return NULL; 3045 if (num_xfrms <= 0) 3046 goto make_dummy_bundle; 3047 3048 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, 3049 xflo->dst_orig); 3050 if (IS_ERR(xdst)) { 3051 err = PTR_ERR(xdst); 3052 if (err == -EREMOTE) { 3053 xfrm_pols_put(pols, num_pols); 3054 return NULL; 3055 } 3056 3057 if (err != -EAGAIN) 3058 goto error; 3059 goto make_dummy_bundle; 3060 } else if (xdst == NULL) { 3061 num_xfrms = 0; 3062 goto make_dummy_bundle; 3063 } 3064 3065 return xdst; 3066 3067 make_dummy_bundle: 3068 /* We found policies, but there's no bundles to instantiate: 3069 * either because the policy blocks, has no transformations or 3070 * we could not build template (no xfrm_states).*/ 3071 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family); 3072 if (IS_ERR(xdst)) { 3073 xfrm_pols_put(pols, num_pols); 3074 return ERR_CAST(xdst); 3075 } 3076 xdst->num_pols = num_pols; 3077 xdst->num_xfrms = num_xfrms; 3078 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols); 3079 3080 return xdst; 3081 3082 inc_error: 3083 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR); 3084 error: 3085 xfrm_pols_put(pols, num_pols); 3086 return ERR_PTR(err); 3087 } 3088 3089 static struct dst_entry *make_blackhole(struct net *net, u16 family, 3090 struct dst_entry *dst_orig) 3091 { 3092 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 3093 struct dst_entry *ret; 3094 3095 if (!afinfo) { 3096 dst_release(dst_orig); 3097 return ERR_PTR(-EINVAL); 3098 } else { 3099 ret = afinfo->blackhole_route(net, dst_orig); 3100 } 3101 rcu_read_unlock(); 3102 3103 return ret; 3104 } 3105 3106 /* Finds/creates a bundle for given flow and if_id 3107 * 3108 * At the moment we eat a raw IP route. Mostly to speed up lookups 3109 * on interfaces with disabled IPsec. 3110 * 3111 * xfrm_lookup uses an if_id of 0 by default, and is provided for 3112 * compatibility 3113 */ 3114 struct dst_entry *xfrm_lookup_with_ifid(struct net *net, 3115 struct dst_entry *dst_orig, 3116 const struct flowi *fl, 3117 const struct sock *sk, 3118 int flags, u32 if_id) 3119 { 3120 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 3121 struct xfrm_dst *xdst; 3122 struct dst_entry *dst, *route; 3123 u16 family = dst_orig->ops->family; 3124 u8 dir = XFRM_POLICY_OUT; 3125 int i, err, num_pols, num_xfrms = 0, drop_pols = 0; 3126 3127 dst = NULL; 3128 xdst = NULL; 3129 route = NULL; 3130 3131 sk = sk_const_to_full_sk(sk); 3132 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) { 3133 num_pols = 1; 3134 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family, 3135 if_id); 3136 err = xfrm_expand_policies(fl, family, pols, 3137 &num_pols, &num_xfrms); 3138 if (err < 0) 3139 goto dropdst; 3140 3141 if (num_pols) { 3142 if (num_xfrms <= 0) { 3143 drop_pols = num_pols; 3144 goto no_transform; 3145 } 3146 3147 xdst = xfrm_resolve_and_create_bundle( 3148 pols, num_pols, fl, 3149 family, dst_orig); 3150 3151 if (IS_ERR(xdst)) { 3152 xfrm_pols_put(pols, num_pols); 3153 err = PTR_ERR(xdst); 3154 if (err == -EREMOTE) 3155 goto nopol; 3156 3157 goto dropdst; 3158 } else if (xdst == NULL) { 3159 num_xfrms = 0; 3160 drop_pols = num_pols; 3161 goto no_transform; 3162 } 3163 3164 route = xdst->route; 3165 } 3166 } 3167 3168 if (xdst == NULL) { 3169 struct xfrm_flo xflo; 3170 3171 xflo.dst_orig = dst_orig; 3172 xflo.flags = flags; 3173 3174 /* To accelerate a bit... */ 3175 if (!if_id && ((dst_orig->flags & DST_NOXFRM) || 3176 !net->xfrm.policy_count[XFRM_POLICY_OUT])) 3177 goto nopol; 3178 3179 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id); 3180 if (xdst == NULL) 3181 goto nopol; 3182 if (IS_ERR(xdst)) { 3183 err = PTR_ERR(xdst); 3184 goto dropdst; 3185 } 3186 3187 num_pols = xdst->num_pols; 3188 num_xfrms = xdst->num_xfrms; 3189 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols); 3190 route = xdst->route; 3191 } 3192 3193 dst = &xdst->u.dst; 3194 if (route == NULL && num_xfrms > 0) { 3195 /* The only case when xfrm_bundle_lookup() returns a 3196 * bundle with null route, is when the template could 3197 * not be resolved. It means policies are there, but 3198 * bundle could not be created, since we don't yet 3199 * have the xfrm_state's. We need to wait for KM to 3200 * negotiate new SA's or bail out with error.*/ 3201 if (net->xfrm.sysctl_larval_drop) { 3202 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES); 3203 err = -EREMOTE; 3204 goto error; 3205 } 3206 3207 err = -EAGAIN; 3208 3209 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES); 3210 goto error; 3211 } 3212 3213 no_transform: 3214 if (num_pols == 0) 3215 goto nopol; 3216 3217 if ((flags & XFRM_LOOKUP_ICMP) && 3218 !(pols[0]->flags & XFRM_POLICY_ICMP)) { 3219 err = -ENOENT; 3220 goto error; 3221 } 3222 3223 for (i = 0; i < num_pols; i++) 3224 WRITE_ONCE(pols[i]->curlft.use_time, ktime_get_real_seconds()); 3225 3226 if (num_xfrms < 0) { 3227 /* Prohibit the flow */ 3228 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK); 3229 err = -EPERM; 3230 goto error; 3231 } else if (num_xfrms > 0) { 3232 /* Flow transformed */ 3233 dst_release(dst_orig); 3234 } else { 3235 /* Flow passes untransformed */ 3236 dst_release(dst); 3237 dst = dst_orig; 3238 } 3239 ok: 3240 xfrm_pols_put(pols, drop_pols); 3241 if (dst && dst->xfrm && 3242 dst->xfrm->props.mode == XFRM_MODE_TUNNEL) 3243 dst->flags |= DST_XFRM_TUNNEL; 3244 return dst; 3245 3246 nopol: 3247 if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) && 3248 net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) { 3249 err = -EPERM; 3250 goto error; 3251 } 3252 if (!(flags & XFRM_LOOKUP_ICMP)) { 3253 dst = dst_orig; 3254 goto ok; 3255 } 3256 err = -ENOENT; 3257 error: 3258 dst_release(dst); 3259 dropdst: 3260 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF)) 3261 dst_release(dst_orig); 3262 xfrm_pols_put(pols, drop_pols); 3263 return ERR_PTR(err); 3264 } 3265 EXPORT_SYMBOL(xfrm_lookup_with_ifid); 3266 3267 /* Main function: finds/creates a bundle for given flow. 3268 * 3269 * At the moment we eat a raw IP route. Mostly to speed up lookups 3270 * on interfaces with disabled IPsec. 3271 */ 3272 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig, 3273 const struct flowi *fl, const struct sock *sk, 3274 int flags) 3275 { 3276 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0); 3277 } 3278 EXPORT_SYMBOL(xfrm_lookup); 3279 3280 /* Callers of xfrm_lookup_route() must ensure a call to dst_output(). 3281 * Otherwise we may send out blackholed packets. 3282 */ 3283 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig, 3284 const struct flowi *fl, 3285 const struct sock *sk, int flags) 3286 { 3287 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk, 3288 flags | XFRM_LOOKUP_QUEUE | 3289 XFRM_LOOKUP_KEEP_DST_REF); 3290 3291 if (PTR_ERR(dst) == -EREMOTE) 3292 return make_blackhole(net, dst_orig->ops->family, dst_orig); 3293 3294 if (IS_ERR(dst)) 3295 dst_release(dst_orig); 3296 3297 return dst; 3298 } 3299 EXPORT_SYMBOL(xfrm_lookup_route); 3300 3301 static inline int 3302 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl) 3303 { 3304 struct sec_path *sp = skb_sec_path(skb); 3305 struct xfrm_state *x; 3306 3307 if (!sp || idx < 0 || idx >= sp->len) 3308 return 0; 3309 x = sp->xvec[idx]; 3310 if (!x->type->reject) 3311 return 0; 3312 return x->type->reject(x, skb, fl); 3313 } 3314 3315 /* When skb is transformed back to its "native" form, we have to 3316 * check policy restrictions. At the moment we make this in maximally 3317 * stupid way. Shame on me. :-) Of course, connected sockets must 3318 * have policy cached at them. 3319 */ 3320 3321 static inline int 3322 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x, 3323 unsigned short family, u32 if_id) 3324 { 3325 if (xfrm_state_kern(x)) 3326 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family); 3327 return x->id.proto == tmpl->id.proto && 3328 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) && 3329 (x->props.reqid == tmpl->reqid || !tmpl->reqid) && 3330 x->props.mode == tmpl->mode && 3331 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) || 3332 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) && 3333 !(x->props.mode != XFRM_MODE_TRANSPORT && 3334 xfrm_state_addr_cmp(tmpl, x, family)) && 3335 (if_id == 0 || if_id == x->if_id); 3336 } 3337 3338 /* 3339 * 0 or more than 0 is returned when validation is succeeded (either bypass 3340 * because of optional transport mode, or next index of the matched secpath 3341 * state with the template. 3342 * -1 is returned when no matching template is found. 3343 * Otherwise "-2 - errored_index" is returned. 3344 */ 3345 static inline int 3346 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start, 3347 unsigned short family, u32 if_id) 3348 { 3349 int idx = start; 3350 3351 if (tmpl->optional) { 3352 if (tmpl->mode == XFRM_MODE_TRANSPORT) 3353 return start; 3354 } else 3355 start = -1; 3356 for (; idx < sp->len; idx++) { 3357 if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id)) 3358 return ++idx; 3359 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) { 3360 if (idx < sp->verified_cnt) { 3361 /* Secpath entry previously verified, consider optional and 3362 * continue searching 3363 */ 3364 continue; 3365 } 3366 3367 if (start == -1) 3368 start = -2-idx; 3369 break; 3370 } 3371 } 3372 return start; 3373 } 3374 3375 static void 3376 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse) 3377 { 3378 const struct iphdr *iph = ip_hdr(skb); 3379 int ihl = iph->ihl; 3380 u8 *xprth = skb_network_header(skb) + ihl * 4; 3381 struct flowi4 *fl4 = &fl->u.ip4; 3382 int oif = 0; 3383 3384 if (skb_dst(skb) && skb_dst(skb)->dev) 3385 oif = skb_dst(skb)->dev->ifindex; 3386 3387 memset(fl4, 0, sizeof(struct flowi4)); 3388 fl4->flowi4_mark = skb->mark; 3389 fl4->flowi4_oif = reverse ? skb->skb_iif : oif; 3390 3391 fl4->flowi4_proto = iph->protocol; 3392 fl4->daddr = reverse ? iph->saddr : iph->daddr; 3393 fl4->saddr = reverse ? iph->daddr : iph->saddr; 3394 fl4->flowi4_tos = iph->tos & ~INET_ECN_MASK; 3395 3396 if (!ip_is_fragment(iph)) { 3397 switch (iph->protocol) { 3398 case IPPROTO_UDP: 3399 case IPPROTO_UDPLITE: 3400 case IPPROTO_TCP: 3401 case IPPROTO_SCTP: 3402 case IPPROTO_DCCP: 3403 if (xprth + 4 < skb->data || 3404 pskb_may_pull(skb, xprth + 4 - skb->data)) { 3405 __be16 *ports; 3406 3407 xprth = skb_network_header(skb) + ihl * 4; 3408 ports = (__be16 *)xprth; 3409 3410 fl4->fl4_sport = ports[!!reverse]; 3411 fl4->fl4_dport = ports[!reverse]; 3412 } 3413 break; 3414 case IPPROTO_ICMP: 3415 if (xprth + 2 < skb->data || 3416 pskb_may_pull(skb, xprth + 2 - skb->data)) { 3417 u8 *icmp; 3418 3419 xprth = skb_network_header(skb) + ihl * 4; 3420 icmp = xprth; 3421 3422 fl4->fl4_icmp_type = icmp[0]; 3423 fl4->fl4_icmp_code = icmp[1]; 3424 } 3425 break; 3426 case IPPROTO_GRE: 3427 if (xprth + 12 < skb->data || 3428 pskb_may_pull(skb, xprth + 12 - skb->data)) { 3429 __be16 *greflags; 3430 __be32 *gre_hdr; 3431 3432 xprth = skb_network_header(skb) + ihl * 4; 3433 greflags = (__be16 *)xprth; 3434 gre_hdr = (__be32 *)xprth; 3435 3436 if (greflags[0] & GRE_KEY) { 3437 if (greflags[0] & GRE_CSUM) 3438 gre_hdr++; 3439 fl4->fl4_gre_key = gre_hdr[1]; 3440 } 3441 } 3442 break; 3443 default: 3444 break; 3445 } 3446 } 3447 } 3448 3449 #if IS_ENABLED(CONFIG_IPV6) 3450 static void 3451 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse) 3452 { 3453 struct flowi6 *fl6 = &fl->u.ip6; 3454 int onlyproto = 0; 3455 const struct ipv6hdr *hdr = ipv6_hdr(skb); 3456 u32 offset = sizeof(*hdr); 3457 struct ipv6_opt_hdr *exthdr; 3458 const unsigned char *nh = skb_network_header(skb); 3459 u16 nhoff = IP6CB(skb)->nhoff; 3460 int oif = 0; 3461 u8 nexthdr; 3462 3463 if (!nhoff) 3464 nhoff = offsetof(struct ipv6hdr, nexthdr); 3465 3466 nexthdr = nh[nhoff]; 3467 3468 if (skb_dst(skb) && skb_dst(skb)->dev) 3469 oif = skb_dst(skb)->dev->ifindex; 3470 3471 memset(fl6, 0, sizeof(struct flowi6)); 3472 fl6->flowi6_mark = skb->mark; 3473 fl6->flowi6_oif = reverse ? skb->skb_iif : oif; 3474 3475 fl6->daddr = reverse ? hdr->saddr : hdr->daddr; 3476 fl6->saddr = reverse ? hdr->daddr : hdr->saddr; 3477 3478 while (nh + offset + sizeof(*exthdr) < skb->data || 3479 pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) { 3480 nh = skb_network_header(skb); 3481 exthdr = (struct ipv6_opt_hdr *)(nh + offset); 3482 3483 switch (nexthdr) { 3484 case NEXTHDR_FRAGMENT: 3485 onlyproto = 1; 3486 fallthrough; 3487 case NEXTHDR_ROUTING: 3488 case NEXTHDR_HOP: 3489 case NEXTHDR_DEST: 3490 offset += ipv6_optlen(exthdr); 3491 nexthdr = exthdr->nexthdr; 3492 break; 3493 case IPPROTO_UDP: 3494 case IPPROTO_UDPLITE: 3495 case IPPROTO_TCP: 3496 case IPPROTO_SCTP: 3497 case IPPROTO_DCCP: 3498 if (!onlyproto && (nh + offset + 4 < skb->data || 3499 pskb_may_pull(skb, nh + offset + 4 - skb->data))) { 3500 __be16 *ports; 3501 3502 nh = skb_network_header(skb); 3503 ports = (__be16 *)(nh + offset); 3504 fl6->fl6_sport = ports[!!reverse]; 3505 fl6->fl6_dport = ports[!reverse]; 3506 } 3507 fl6->flowi6_proto = nexthdr; 3508 return; 3509 case IPPROTO_ICMPV6: 3510 if (!onlyproto && (nh + offset + 2 < skb->data || 3511 pskb_may_pull(skb, nh + offset + 2 - skb->data))) { 3512 u8 *icmp; 3513 3514 nh = skb_network_header(skb); 3515 icmp = (u8 *)(nh + offset); 3516 fl6->fl6_icmp_type = icmp[0]; 3517 fl6->fl6_icmp_code = icmp[1]; 3518 } 3519 fl6->flowi6_proto = nexthdr; 3520 return; 3521 case IPPROTO_GRE: 3522 if (!onlyproto && 3523 (nh + offset + 12 < skb->data || 3524 pskb_may_pull(skb, nh + offset + 12 - skb->data))) { 3525 struct gre_base_hdr *gre_hdr; 3526 __be32 *gre_key; 3527 3528 nh = skb_network_header(skb); 3529 gre_hdr = (struct gre_base_hdr *)(nh + offset); 3530 gre_key = (__be32 *)(gre_hdr + 1); 3531 3532 if (gre_hdr->flags & GRE_KEY) { 3533 if (gre_hdr->flags & GRE_CSUM) 3534 gre_key++; 3535 fl6->fl6_gre_key = *gre_key; 3536 } 3537 } 3538 fl6->flowi6_proto = nexthdr; 3539 return; 3540 3541 #if IS_ENABLED(CONFIG_IPV6_MIP6) 3542 case IPPROTO_MH: 3543 offset += ipv6_optlen(exthdr); 3544 if (!onlyproto && (nh + offset + 3 < skb->data || 3545 pskb_may_pull(skb, nh + offset + 3 - skb->data))) { 3546 struct ip6_mh *mh; 3547 3548 nh = skb_network_header(skb); 3549 mh = (struct ip6_mh *)(nh + offset); 3550 fl6->fl6_mh_type = mh->ip6mh_type; 3551 } 3552 fl6->flowi6_proto = nexthdr; 3553 return; 3554 #endif 3555 default: 3556 fl6->flowi6_proto = nexthdr; 3557 return; 3558 } 3559 } 3560 } 3561 #endif 3562 3563 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, 3564 unsigned int family, int reverse) 3565 { 3566 switch (family) { 3567 case AF_INET: 3568 decode_session4(skb, fl, reverse); 3569 break; 3570 #if IS_ENABLED(CONFIG_IPV6) 3571 case AF_INET6: 3572 decode_session6(skb, fl, reverse); 3573 break; 3574 #endif 3575 default: 3576 return -EAFNOSUPPORT; 3577 } 3578 3579 return security_xfrm_decode_session(skb, &fl->flowi_secid); 3580 } 3581 EXPORT_SYMBOL(__xfrm_decode_session); 3582 3583 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp) 3584 { 3585 for (; k < sp->len; k++) { 3586 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) { 3587 *idxp = k; 3588 return 1; 3589 } 3590 } 3591 3592 return 0; 3593 } 3594 3595 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 3596 unsigned short family) 3597 { 3598 struct net *net = dev_net(skb->dev); 3599 struct xfrm_policy *pol; 3600 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 3601 int npols = 0; 3602 int xfrm_nr; 3603 int pi; 3604 int reverse; 3605 struct flowi fl; 3606 int xerr_idx = -1; 3607 const struct xfrm_if_cb *ifcb; 3608 struct sec_path *sp; 3609 u32 if_id = 0; 3610 3611 rcu_read_lock(); 3612 ifcb = xfrm_if_get_cb(); 3613 3614 if (ifcb) { 3615 struct xfrm_if_decode_session_result r; 3616 3617 if (ifcb->decode_session(skb, family, &r)) { 3618 if_id = r.if_id; 3619 net = r.net; 3620 } 3621 } 3622 rcu_read_unlock(); 3623 3624 reverse = dir & ~XFRM_POLICY_MASK; 3625 dir &= XFRM_POLICY_MASK; 3626 3627 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) { 3628 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR); 3629 return 0; 3630 } 3631 3632 nf_nat_decode_session(skb, &fl, family); 3633 3634 /* First, check used SA against their selectors. */ 3635 sp = skb_sec_path(skb); 3636 if (sp) { 3637 int i; 3638 3639 for (i = sp->len - 1; i >= 0; i--) { 3640 struct xfrm_state *x = sp->xvec[i]; 3641 if (!xfrm_selector_match(&x->sel, &fl, family)) { 3642 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH); 3643 return 0; 3644 } 3645 } 3646 } 3647 3648 pol = NULL; 3649 sk = sk_to_full_sk(sk); 3650 if (sk && sk->sk_policy[dir]) { 3651 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id); 3652 if (IS_ERR(pol)) { 3653 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3654 return 0; 3655 } 3656 } 3657 3658 if (!pol) 3659 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id); 3660 3661 if (IS_ERR(pol)) { 3662 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3663 return 0; 3664 } 3665 3666 if (!pol) { 3667 if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) { 3668 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS); 3669 return 0; 3670 } 3671 3672 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) { 3673 xfrm_secpath_reject(xerr_idx, skb, &fl); 3674 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS); 3675 return 0; 3676 } 3677 return 1; 3678 } 3679 3680 /* This lockless write can happen from different cpus. */ 3681 WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds()); 3682 3683 pols[0] = pol; 3684 npols++; 3685 #ifdef CONFIG_XFRM_SUB_POLICY 3686 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) { 3687 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, 3688 &fl, family, 3689 XFRM_POLICY_IN, if_id); 3690 if (pols[1]) { 3691 if (IS_ERR(pols[1])) { 3692 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3693 xfrm_pol_put(pols[0]); 3694 return 0; 3695 } 3696 /* This write can happen from different cpus. */ 3697 WRITE_ONCE(pols[1]->curlft.use_time, 3698 ktime_get_real_seconds()); 3699 npols++; 3700 } 3701 } 3702 #endif 3703 3704 if (pol->action == XFRM_POLICY_ALLOW) { 3705 static struct sec_path dummy; 3706 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH]; 3707 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH]; 3708 struct xfrm_tmpl **tpp = tp; 3709 int ti = 0; 3710 int i, k; 3711 3712 sp = skb_sec_path(skb); 3713 if (!sp) 3714 sp = &dummy; 3715 3716 for (pi = 0; pi < npols; pi++) { 3717 if (pols[pi] != pol && 3718 pols[pi]->action != XFRM_POLICY_ALLOW) { 3719 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK); 3720 goto reject; 3721 } 3722 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) { 3723 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR); 3724 goto reject_error; 3725 } 3726 for (i = 0; i < pols[pi]->xfrm_nr; i++) 3727 tpp[ti++] = &pols[pi]->xfrm_vec[i]; 3728 } 3729 xfrm_nr = ti; 3730 3731 if (npols > 1) { 3732 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family); 3733 tpp = stp; 3734 } 3735 3736 /* For each tunnel xfrm, find the first matching tmpl. 3737 * For each tmpl before that, find corresponding xfrm. 3738 * Order is _important_. Later we will implement 3739 * some barriers, but at the moment barriers 3740 * are implied between each two transformations. 3741 * Upon success, marks secpath entries as having been 3742 * verified to allow them to be skipped in future policy 3743 * checks (e.g. nested tunnels). 3744 */ 3745 for (i = xfrm_nr-1, k = 0; i >= 0; i--) { 3746 k = xfrm_policy_ok(tpp[i], sp, k, family, if_id); 3747 if (k < 0) { 3748 if (k < -1) 3749 /* "-2 - errored_index" returned */ 3750 xerr_idx = -(2+k); 3751 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH); 3752 goto reject; 3753 } 3754 } 3755 3756 if (secpath_has_nontransport(sp, k, &xerr_idx)) { 3757 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH); 3758 goto reject; 3759 } 3760 3761 xfrm_pols_put(pols, npols); 3762 sp->verified_cnt = k; 3763 3764 return 1; 3765 } 3766 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK); 3767 3768 reject: 3769 xfrm_secpath_reject(xerr_idx, skb, &fl); 3770 reject_error: 3771 xfrm_pols_put(pols, npols); 3772 return 0; 3773 } 3774 EXPORT_SYMBOL(__xfrm_policy_check); 3775 3776 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family) 3777 { 3778 struct net *net = dev_net(skb->dev); 3779 struct flowi fl; 3780 struct dst_entry *dst; 3781 int res = 1; 3782 3783 if (xfrm_decode_session(skb, &fl, family) < 0) { 3784 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR); 3785 return 0; 3786 } 3787 3788 skb_dst_force(skb); 3789 if (!skb_dst(skb)) { 3790 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR); 3791 return 0; 3792 } 3793 3794 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE); 3795 if (IS_ERR(dst)) { 3796 res = 0; 3797 dst = NULL; 3798 } 3799 skb_dst_set(skb, dst); 3800 return res; 3801 } 3802 EXPORT_SYMBOL(__xfrm_route_forward); 3803 3804 /* Optimize later using cookies and generation ids. */ 3805 3806 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie) 3807 { 3808 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete 3809 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to 3810 * get validated by dst_ops->check on every use. We do this 3811 * because when a normal route referenced by an XFRM dst is 3812 * obsoleted we do not go looking around for all parent 3813 * referencing XFRM dsts so that we can invalidate them. It 3814 * is just too much work. Instead we make the checks here on 3815 * every use. For example: 3816 * 3817 * XFRM dst A --> IPv4 dst X 3818 * 3819 * X is the "xdst->route" of A (X is also the "dst->path" of A 3820 * in this example). If X is marked obsolete, "A" will not 3821 * notice. That's what we are validating here via the 3822 * stale_bundle() check. 3823 * 3824 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will 3825 * be marked on it. 3826 * This will force stale_bundle() to fail on any xdst bundle with 3827 * this dst linked in it. 3828 */ 3829 if (dst->obsolete < 0 && !stale_bundle(dst)) 3830 return dst; 3831 3832 return NULL; 3833 } 3834 3835 static int stale_bundle(struct dst_entry *dst) 3836 { 3837 return !xfrm_bundle_ok((struct xfrm_dst *)dst); 3838 } 3839 3840 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev) 3841 { 3842 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) { 3843 dst->dev = blackhole_netdev; 3844 dev_hold(dst->dev); 3845 dev_put(dev); 3846 } 3847 } 3848 EXPORT_SYMBOL(xfrm_dst_ifdown); 3849 3850 static void xfrm_link_failure(struct sk_buff *skb) 3851 { 3852 /* Impossible. Such dst must be popped before reaches point of failure. */ 3853 } 3854 3855 static void xfrm_negative_advice(struct sock *sk, struct dst_entry *dst) 3856 { 3857 if (dst->obsolete) 3858 sk_dst_reset(sk); 3859 } 3860 3861 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr) 3862 { 3863 while (nr--) { 3864 struct xfrm_dst *xdst = bundle[nr]; 3865 u32 pmtu, route_mtu_cached; 3866 struct dst_entry *dst; 3867 3868 dst = &xdst->u.dst; 3869 pmtu = dst_mtu(xfrm_dst_child(dst)); 3870 xdst->child_mtu_cached = pmtu; 3871 3872 pmtu = xfrm_state_mtu(dst->xfrm, pmtu); 3873 3874 route_mtu_cached = dst_mtu(xdst->route); 3875 xdst->route_mtu_cached = route_mtu_cached; 3876 3877 if (pmtu > route_mtu_cached) 3878 pmtu = route_mtu_cached; 3879 3880 dst_metric_set(dst, RTAX_MTU, pmtu); 3881 } 3882 } 3883 3884 /* Check that the bundle accepts the flow and its components are 3885 * still valid. 3886 */ 3887 3888 static int xfrm_bundle_ok(struct xfrm_dst *first) 3889 { 3890 struct xfrm_dst *bundle[XFRM_MAX_DEPTH]; 3891 struct dst_entry *dst = &first->u.dst; 3892 struct xfrm_dst *xdst; 3893 int start_from, nr; 3894 u32 mtu; 3895 3896 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) || 3897 (dst->dev && !netif_running(dst->dev))) 3898 return 0; 3899 3900 if (dst->flags & DST_XFRM_QUEUE) 3901 return 1; 3902 3903 start_from = nr = 0; 3904 do { 3905 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 3906 3907 if (dst->xfrm->km.state != XFRM_STATE_VALID) 3908 return 0; 3909 if (xdst->xfrm_genid != dst->xfrm->genid) 3910 return 0; 3911 if (xdst->num_pols > 0 && 3912 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid)) 3913 return 0; 3914 3915 bundle[nr++] = xdst; 3916 3917 mtu = dst_mtu(xfrm_dst_child(dst)); 3918 if (xdst->child_mtu_cached != mtu) { 3919 start_from = nr; 3920 xdst->child_mtu_cached = mtu; 3921 } 3922 3923 if (!dst_check(xdst->route, xdst->route_cookie)) 3924 return 0; 3925 mtu = dst_mtu(xdst->route); 3926 if (xdst->route_mtu_cached != mtu) { 3927 start_from = nr; 3928 xdst->route_mtu_cached = mtu; 3929 } 3930 3931 dst = xfrm_dst_child(dst); 3932 } while (dst->xfrm); 3933 3934 if (likely(!start_from)) 3935 return 1; 3936 3937 xdst = bundle[start_from - 1]; 3938 mtu = xdst->child_mtu_cached; 3939 while (start_from--) { 3940 dst = &xdst->u.dst; 3941 3942 mtu = xfrm_state_mtu(dst->xfrm, mtu); 3943 if (mtu > xdst->route_mtu_cached) 3944 mtu = xdst->route_mtu_cached; 3945 dst_metric_set(dst, RTAX_MTU, mtu); 3946 if (!start_from) 3947 break; 3948 3949 xdst = bundle[start_from - 1]; 3950 xdst->child_mtu_cached = mtu; 3951 } 3952 3953 return 1; 3954 } 3955 3956 static unsigned int xfrm_default_advmss(const struct dst_entry *dst) 3957 { 3958 return dst_metric_advmss(xfrm_dst_path(dst)); 3959 } 3960 3961 static unsigned int xfrm_mtu(const struct dst_entry *dst) 3962 { 3963 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 3964 3965 return mtu ? : dst_mtu(xfrm_dst_path(dst)); 3966 } 3967 3968 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst, 3969 const void *daddr) 3970 { 3971 while (dst->xfrm) { 3972 const struct xfrm_state *xfrm = dst->xfrm; 3973 3974 dst = xfrm_dst_child(dst); 3975 3976 if (xfrm->props.mode == XFRM_MODE_TRANSPORT) 3977 continue; 3978 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR) 3979 daddr = xfrm->coaddr; 3980 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR)) 3981 daddr = &xfrm->id.daddr; 3982 } 3983 return daddr; 3984 } 3985 3986 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst, 3987 struct sk_buff *skb, 3988 const void *daddr) 3989 { 3990 const struct dst_entry *path = xfrm_dst_path(dst); 3991 3992 if (!skb) 3993 daddr = xfrm_get_dst_nexthop(dst, daddr); 3994 return path->ops->neigh_lookup(path, skb, daddr); 3995 } 3996 3997 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr) 3998 { 3999 const struct dst_entry *path = xfrm_dst_path(dst); 4000 4001 daddr = xfrm_get_dst_nexthop(dst, daddr); 4002 path->ops->confirm_neigh(path, daddr); 4003 } 4004 4005 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family) 4006 { 4007 int err = 0; 4008 4009 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo))) 4010 return -EAFNOSUPPORT; 4011 4012 spin_lock(&xfrm_policy_afinfo_lock); 4013 if (unlikely(xfrm_policy_afinfo[family] != NULL)) 4014 err = -EEXIST; 4015 else { 4016 struct dst_ops *dst_ops = afinfo->dst_ops; 4017 if (likely(dst_ops->kmem_cachep == NULL)) 4018 dst_ops->kmem_cachep = xfrm_dst_cache; 4019 if (likely(dst_ops->check == NULL)) 4020 dst_ops->check = xfrm_dst_check; 4021 if (likely(dst_ops->default_advmss == NULL)) 4022 dst_ops->default_advmss = xfrm_default_advmss; 4023 if (likely(dst_ops->mtu == NULL)) 4024 dst_ops->mtu = xfrm_mtu; 4025 if (likely(dst_ops->negative_advice == NULL)) 4026 dst_ops->negative_advice = xfrm_negative_advice; 4027 if (likely(dst_ops->link_failure == NULL)) 4028 dst_ops->link_failure = xfrm_link_failure; 4029 if (likely(dst_ops->neigh_lookup == NULL)) 4030 dst_ops->neigh_lookup = xfrm_neigh_lookup; 4031 if (likely(!dst_ops->confirm_neigh)) 4032 dst_ops->confirm_neigh = xfrm_confirm_neigh; 4033 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo); 4034 } 4035 spin_unlock(&xfrm_policy_afinfo_lock); 4036 4037 return err; 4038 } 4039 EXPORT_SYMBOL(xfrm_policy_register_afinfo); 4040 4041 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo) 4042 { 4043 struct dst_ops *dst_ops = afinfo->dst_ops; 4044 int i; 4045 4046 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) { 4047 if (xfrm_policy_afinfo[i] != afinfo) 4048 continue; 4049 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL); 4050 break; 4051 } 4052 4053 synchronize_rcu(); 4054 4055 dst_ops->kmem_cachep = NULL; 4056 dst_ops->check = NULL; 4057 dst_ops->negative_advice = NULL; 4058 dst_ops->link_failure = NULL; 4059 } 4060 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo); 4061 4062 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb) 4063 { 4064 spin_lock(&xfrm_if_cb_lock); 4065 rcu_assign_pointer(xfrm_if_cb, ifcb); 4066 spin_unlock(&xfrm_if_cb_lock); 4067 } 4068 EXPORT_SYMBOL(xfrm_if_register_cb); 4069 4070 void xfrm_if_unregister_cb(void) 4071 { 4072 RCU_INIT_POINTER(xfrm_if_cb, NULL); 4073 synchronize_rcu(); 4074 } 4075 EXPORT_SYMBOL(xfrm_if_unregister_cb); 4076 4077 #ifdef CONFIG_XFRM_STATISTICS 4078 static int __net_init xfrm_statistics_init(struct net *net) 4079 { 4080 int rv; 4081 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib); 4082 if (!net->mib.xfrm_statistics) 4083 return -ENOMEM; 4084 rv = xfrm_proc_init(net); 4085 if (rv < 0) 4086 free_percpu(net->mib.xfrm_statistics); 4087 return rv; 4088 } 4089 4090 static void xfrm_statistics_fini(struct net *net) 4091 { 4092 xfrm_proc_fini(net); 4093 free_percpu(net->mib.xfrm_statistics); 4094 } 4095 #else 4096 static int __net_init xfrm_statistics_init(struct net *net) 4097 { 4098 return 0; 4099 } 4100 4101 static void xfrm_statistics_fini(struct net *net) 4102 { 4103 } 4104 #endif 4105 4106 static int __net_init xfrm_policy_init(struct net *net) 4107 { 4108 unsigned int hmask, sz; 4109 int dir, err; 4110 4111 if (net_eq(net, &init_net)) { 4112 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache", 4113 sizeof(struct xfrm_dst), 4114 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4115 NULL); 4116 err = rhashtable_init(&xfrm_policy_inexact_table, 4117 &xfrm_pol_inexact_params); 4118 BUG_ON(err); 4119 } 4120 4121 hmask = 8 - 1; 4122 sz = (hmask+1) * sizeof(struct hlist_head); 4123 4124 net->xfrm.policy_byidx = xfrm_hash_alloc(sz); 4125 if (!net->xfrm.policy_byidx) 4126 goto out_byidx; 4127 net->xfrm.policy_idx_hmask = hmask; 4128 4129 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 4130 struct xfrm_policy_hash *htab; 4131 4132 net->xfrm.policy_count[dir] = 0; 4133 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0; 4134 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]); 4135 4136 htab = &net->xfrm.policy_bydst[dir]; 4137 htab->table = xfrm_hash_alloc(sz); 4138 if (!htab->table) 4139 goto out_bydst; 4140 htab->hmask = hmask; 4141 htab->dbits4 = 32; 4142 htab->sbits4 = 32; 4143 htab->dbits6 = 128; 4144 htab->sbits6 = 128; 4145 } 4146 net->xfrm.policy_hthresh.lbits4 = 32; 4147 net->xfrm.policy_hthresh.rbits4 = 32; 4148 net->xfrm.policy_hthresh.lbits6 = 128; 4149 net->xfrm.policy_hthresh.rbits6 = 128; 4150 4151 seqlock_init(&net->xfrm.policy_hthresh.lock); 4152 4153 INIT_LIST_HEAD(&net->xfrm.policy_all); 4154 INIT_LIST_HEAD(&net->xfrm.inexact_bins); 4155 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize); 4156 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild); 4157 return 0; 4158 4159 out_bydst: 4160 for (dir--; dir >= 0; dir--) { 4161 struct xfrm_policy_hash *htab; 4162 4163 htab = &net->xfrm.policy_bydst[dir]; 4164 xfrm_hash_free(htab->table, sz); 4165 } 4166 xfrm_hash_free(net->xfrm.policy_byidx, sz); 4167 out_byidx: 4168 return -ENOMEM; 4169 } 4170 4171 static void xfrm_policy_fini(struct net *net) 4172 { 4173 struct xfrm_pol_inexact_bin *b, *t; 4174 unsigned int sz; 4175 int dir; 4176 4177 flush_work(&net->xfrm.policy_hash_work); 4178 #ifdef CONFIG_XFRM_SUB_POLICY 4179 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false); 4180 #endif 4181 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false); 4182 4183 WARN_ON(!list_empty(&net->xfrm.policy_all)); 4184 4185 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 4186 struct xfrm_policy_hash *htab; 4187 4188 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir])); 4189 4190 htab = &net->xfrm.policy_bydst[dir]; 4191 sz = (htab->hmask + 1) * sizeof(struct hlist_head); 4192 WARN_ON(!hlist_empty(htab->table)); 4193 xfrm_hash_free(htab->table, sz); 4194 } 4195 4196 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head); 4197 WARN_ON(!hlist_empty(net->xfrm.policy_byidx)); 4198 xfrm_hash_free(net->xfrm.policy_byidx, sz); 4199 4200 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 4201 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins) 4202 __xfrm_policy_inexact_prune_bin(b, true); 4203 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 4204 } 4205 4206 static int __net_init xfrm_net_init(struct net *net) 4207 { 4208 int rv; 4209 4210 /* Initialize the per-net locks here */ 4211 spin_lock_init(&net->xfrm.xfrm_state_lock); 4212 spin_lock_init(&net->xfrm.xfrm_policy_lock); 4213 seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock); 4214 mutex_init(&net->xfrm.xfrm_cfg_mutex); 4215 net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT; 4216 net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT; 4217 net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT; 4218 4219 rv = xfrm_statistics_init(net); 4220 if (rv < 0) 4221 goto out_statistics; 4222 rv = xfrm_state_init(net); 4223 if (rv < 0) 4224 goto out_state; 4225 rv = xfrm_policy_init(net); 4226 if (rv < 0) 4227 goto out_policy; 4228 rv = xfrm_sysctl_init(net); 4229 if (rv < 0) 4230 goto out_sysctl; 4231 4232 return 0; 4233 4234 out_sysctl: 4235 xfrm_policy_fini(net); 4236 out_policy: 4237 xfrm_state_fini(net); 4238 out_state: 4239 xfrm_statistics_fini(net); 4240 out_statistics: 4241 return rv; 4242 } 4243 4244 static void __net_exit xfrm_net_exit(struct net *net) 4245 { 4246 xfrm_sysctl_fini(net); 4247 xfrm_policy_fini(net); 4248 xfrm_state_fini(net); 4249 xfrm_statistics_fini(net); 4250 } 4251 4252 static struct pernet_operations __net_initdata xfrm_net_ops = { 4253 .init = xfrm_net_init, 4254 .exit = xfrm_net_exit, 4255 }; 4256 4257 void __init xfrm_init(void) 4258 { 4259 register_pernet_subsys(&xfrm_net_ops); 4260 xfrm_dev_init(); 4261 xfrm_input_init(); 4262 4263 #ifdef CONFIG_XFRM_ESPINTCP 4264 espintcp_init(); 4265 #endif 4266 } 4267 4268 #ifdef CONFIG_AUDITSYSCALL 4269 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp, 4270 struct audit_buffer *audit_buf) 4271 { 4272 struct xfrm_sec_ctx *ctx = xp->security; 4273 struct xfrm_selector *sel = &xp->selector; 4274 4275 if (ctx) 4276 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s", 4277 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str); 4278 4279 switch (sel->family) { 4280 case AF_INET: 4281 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4); 4282 if (sel->prefixlen_s != 32) 4283 audit_log_format(audit_buf, " src_prefixlen=%d", 4284 sel->prefixlen_s); 4285 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4); 4286 if (sel->prefixlen_d != 32) 4287 audit_log_format(audit_buf, " dst_prefixlen=%d", 4288 sel->prefixlen_d); 4289 break; 4290 case AF_INET6: 4291 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6); 4292 if (sel->prefixlen_s != 128) 4293 audit_log_format(audit_buf, " src_prefixlen=%d", 4294 sel->prefixlen_s); 4295 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6); 4296 if (sel->prefixlen_d != 128) 4297 audit_log_format(audit_buf, " dst_prefixlen=%d", 4298 sel->prefixlen_d); 4299 break; 4300 } 4301 } 4302 4303 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid) 4304 { 4305 struct audit_buffer *audit_buf; 4306 4307 audit_buf = xfrm_audit_start("SPD-add"); 4308 if (audit_buf == NULL) 4309 return; 4310 xfrm_audit_helper_usrinfo(task_valid, audit_buf); 4311 audit_log_format(audit_buf, " res=%u", result); 4312 xfrm_audit_common_policyinfo(xp, audit_buf); 4313 audit_log_end(audit_buf); 4314 } 4315 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add); 4316 4317 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result, 4318 bool task_valid) 4319 { 4320 struct audit_buffer *audit_buf; 4321 4322 audit_buf = xfrm_audit_start("SPD-delete"); 4323 if (audit_buf == NULL) 4324 return; 4325 xfrm_audit_helper_usrinfo(task_valid, audit_buf); 4326 audit_log_format(audit_buf, " res=%u", result); 4327 xfrm_audit_common_policyinfo(xp, audit_buf); 4328 audit_log_end(audit_buf); 4329 } 4330 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete); 4331 #endif 4332 4333 #ifdef CONFIG_XFRM_MIGRATE 4334 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp, 4335 const struct xfrm_selector *sel_tgt) 4336 { 4337 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) { 4338 if (sel_tgt->family == sel_cmp->family && 4339 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr, 4340 sel_cmp->family) && 4341 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr, 4342 sel_cmp->family) && 4343 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d && 4344 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) { 4345 return true; 4346 } 4347 } else { 4348 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) { 4349 return true; 4350 } 4351 } 4352 return false; 4353 } 4354 4355 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel, 4356 u8 dir, u8 type, struct net *net, u32 if_id) 4357 { 4358 struct xfrm_policy *pol, *ret = NULL; 4359 struct hlist_head *chain; 4360 u32 priority = ~0U; 4361 4362 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 4363 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir); 4364 hlist_for_each_entry(pol, chain, bydst) { 4365 if ((if_id == 0 || pol->if_id == if_id) && 4366 xfrm_migrate_selector_match(sel, &pol->selector) && 4367 pol->type == type) { 4368 ret = pol; 4369 priority = ret->priority; 4370 break; 4371 } 4372 } 4373 chain = &net->xfrm.policy_inexact[dir]; 4374 hlist_for_each_entry(pol, chain, bydst_inexact_list) { 4375 if ((pol->priority >= priority) && ret) 4376 break; 4377 4378 if ((if_id == 0 || pol->if_id == if_id) && 4379 xfrm_migrate_selector_match(sel, &pol->selector) && 4380 pol->type == type) { 4381 ret = pol; 4382 break; 4383 } 4384 } 4385 4386 xfrm_pol_hold(ret); 4387 4388 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 4389 4390 return ret; 4391 } 4392 4393 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t) 4394 { 4395 int match = 0; 4396 4397 if (t->mode == m->mode && t->id.proto == m->proto && 4398 (m->reqid == 0 || t->reqid == m->reqid)) { 4399 switch (t->mode) { 4400 case XFRM_MODE_TUNNEL: 4401 case XFRM_MODE_BEET: 4402 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr, 4403 m->old_family) && 4404 xfrm_addr_equal(&t->saddr, &m->old_saddr, 4405 m->old_family)) { 4406 match = 1; 4407 } 4408 break; 4409 case XFRM_MODE_TRANSPORT: 4410 /* in case of transport mode, template does not store 4411 any IP addresses, hence we just compare mode and 4412 protocol */ 4413 match = 1; 4414 break; 4415 default: 4416 break; 4417 } 4418 } 4419 return match; 4420 } 4421 4422 /* update endpoint address(es) of template(s) */ 4423 static int xfrm_policy_migrate(struct xfrm_policy *pol, 4424 struct xfrm_migrate *m, int num_migrate, 4425 struct netlink_ext_ack *extack) 4426 { 4427 struct xfrm_migrate *mp; 4428 int i, j, n = 0; 4429 4430 write_lock_bh(&pol->lock); 4431 if (unlikely(pol->walk.dead)) { 4432 /* target policy has been deleted */ 4433 NL_SET_ERR_MSG(extack, "Target policy not found"); 4434 write_unlock_bh(&pol->lock); 4435 return -ENOENT; 4436 } 4437 4438 for (i = 0; i < pol->xfrm_nr; i++) { 4439 for (j = 0, mp = m; j < num_migrate; j++, mp++) { 4440 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i])) 4441 continue; 4442 n++; 4443 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL && 4444 pol->xfrm_vec[i].mode != XFRM_MODE_BEET) 4445 continue; 4446 /* update endpoints */ 4447 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr, 4448 sizeof(pol->xfrm_vec[i].id.daddr)); 4449 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr, 4450 sizeof(pol->xfrm_vec[i].saddr)); 4451 pol->xfrm_vec[i].encap_family = mp->new_family; 4452 /* flush bundles */ 4453 atomic_inc(&pol->genid); 4454 } 4455 } 4456 4457 write_unlock_bh(&pol->lock); 4458 4459 if (!n) 4460 return -ENODATA; 4461 4462 return 0; 4463 } 4464 4465 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate, 4466 struct netlink_ext_ack *extack) 4467 { 4468 int i, j; 4469 4470 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH) { 4471 NL_SET_ERR_MSG(extack, "Invalid number of SAs to migrate, must be 0 < num <= XFRM_MAX_DEPTH (6)"); 4472 return -EINVAL; 4473 } 4474 4475 for (i = 0; i < num_migrate; i++) { 4476 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) || 4477 xfrm_addr_any(&m[i].new_saddr, m[i].new_family)) { 4478 NL_SET_ERR_MSG(extack, "Addresses in the MIGRATE attribute's list cannot be null"); 4479 return -EINVAL; 4480 } 4481 4482 /* check if there is any duplicated entry */ 4483 for (j = i + 1; j < num_migrate; j++) { 4484 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr, 4485 sizeof(m[i].old_daddr)) && 4486 !memcmp(&m[i].old_saddr, &m[j].old_saddr, 4487 sizeof(m[i].old_saddr)) && 4488 m[i].proto == m[j].proto && 4489 m[i].mode == m[j].mode && 4490 m[i].reqid == m[j].reqid && 4491 m[i].old_family == m[j].old_family) { 4492 NL_SET_ERR_MSG(extack, "Entries in the MIGRATE attribute's list must be unique"); 4493 return -EINVAL; 4494 } 4495 } 4496 } 4497 4498 return 0; 4499 } 4500 4501 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type, 4502 struct xfrm_migrate *m, int num_migrate, 4503 struct xfrm_kmaddress *k, struct net *net, 4504 struct xfrm_encap_tmpl *encap, u32 if_id, 4505 struct netlink_ext_ack *extack) 4506 { 4507 int i, err, nx_cur = 0, nx_new = 0; 4508 struct xfrm_policy *pol = NULL; 4509 struct xfrm_state *x, *xc; 4510 struct xfrm_state *x_cur[XFRM_MAX_DEPTH]; 4511 struct xfrm_state *x_new[XFRM_MAX_DEPTH]; 4512 struct xfrm_migrate *mp; 4513 4514 /* Stage 0 - sanity checks */ 4515 err = xfrm_migrate_check(m, num_migrate, extack); 4516 if (err < 0) 4517 goto out; 4518 4519 if (dir >= XFRM_POLICY_MAX) { 4520 NL_SET_ERR_MSG(extack, "Invalid policy direction"); 4521 err = -EINVAL; 4522 goto out; 4523 } 4524 4525 /* Stage 1 - find policy */ 4526 pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id); 4527 if (!pol) { 4528 NL_SET_ERR_MSG(extack, "Target policy not found"); 4529 err = -ENOENT; 4530 goto out; 4531 } 4532 4533 /* Stage 2 - find and update state(s) */ 4534 for (i = 0, mp = m; i < num_migrate; i++, mp++) { 4535 if ((x = xfrm_migrate_state_find(mp, net, if_id))) { 4536 x_cur[nx_cur] = x; 4537 nx_cur++; 4538 xc = xfrm_state_migrate(x, mp, encap); 4539 if (xc) { 4540 x_new[nx_new] = xc; 4541 nx_new++; 4542 } else { 4543 err = -ENODATA; 4544 goto restore_state; 4545 } 4546 } 4547 } 4548 4549 /* Stage 3 - update policy */ 4550 err = xfrm_policy_migrate(pol, m, num_migrate, extack); 4551 if (err < 0) 4552 goto restore_state; 4553 4554 /* Stage 4 - delete old state(s) */ 4555 if (nx_cur) { 4556 xfrm_states_put(x_cur, nx_cur); 4557 xfrm_states_delete(x_cur, nx_cur); 4558 } 4559 4560 /* Stage 5 - announce */ 4561 km_migrate(sel, dir, type, m, num_migrate, k, encap); 4562 4563 xfrm_pol_put(pol); 4564 4565 return 0; 4566 out: 4567 return err; 4568 4569 restore_state: 4570 if (pol) 4571 xfrm_pol_put(pol); 4572 if (nx_cur) 4573 xfrm_states_put(x_cur, nx_cur); 4574 if (nx_new) 4575 xfrm_states_delete(x_new, nx_new); 4576 4577 return err; 4578 } 4579 EXPORT_SYMBOL(xfrm_migrate); 4580 #endif 4581