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 write_lock_bh(&policy->lock); 440 policy->walk.dead = 1; 441 write_unlock_bh(&policy->lock); 442 443 atomic_inc(&policy->genid); 444 445 if (del_timer(&policy->polq.hold_timer)) 446 xfrm_pol_put(policy); 447 skb_queue_purge(&policy->polq.hold_queue); 448 449 if (del_timer(&policy->timer)) 450 xfrm_pol_put(policy); 451 452 xfrm_pol_put(policy); 453 } 454 455 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024; 456 457 static inline unsigned int idx_hash(struct net *net, u32 index) 458 { 459 return __idx_hash(index, net->xfrm.policy_idx_hmask); 460 } 461 462 /* calculate policy hash thresholds */ 463 static void __get_hash_thresh(struct net *net, 464 unsigned short family, int dir, 465 u8 *dbits, u8 *sbits) 466 { 467 switch (family) { 468 case AF_INET: 469 *dbits = net->xfrm.policy_bydst[dir].dbits4; 470 *sbits = net->xfrm.policy_bydst[dir].sbits4; 471 break; 472 473 case AF_INET6: 474 *dbits = net->xfrm.policy_bydst[dir].dbits6; 475 *sbits = net->xfrm.policy_bydst[dir].sbits6; 476 break; 477 478 default: 479 *dbits = 0; 480 *sbits = 0; 481 } 482 } 483 484 static struct hlist_head *policy_hash_bysel(struct net *net, 485 const struct xfrm_selector *sel, 486 unsigned short family, int dir) 487 { 488 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 489 unsigned int hash; 490 u8 dbits; 491 u8 sbits; 492 493 __get_hash_thresh(net, family, dir, &dbits, &sbits); 494 hash = __sel_hash(sel, family, hmask, dbits, sbits); 495 496 if (hash == hmask + 1) 497 return NULL; 498 499 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table, 500 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash; 501 } 502 503 static struct hlist_head *policy_hash_direct(struct net *net, 504 const xfrm_address_t *daddr, 505 const xfrm_address_t *saddr, 506 unsigned short family, int dir) 507 { 508 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 509 unsigned int hash; 510 u8 dbits; 511 u8 sbits; 512 513 __get_hash_thresh(net, family, dir, &dbits, &sbits); 514 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits); 515 516 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table, 517 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash; 518 } 519 520 static void xfrm_dst_hash_transfer(struct net *net, 521 struct hlist_head *list, 522 struct hlist_head *ndsttable, 523 unsigned int nhashmask, 524 int dir) 525 { 526 struct hlist_node *tmp, *entry0 = NULL; 527 struct xfrm_policy *pol; 528 unsigned int h0 = 0; 529 u8 dbits; 530 u8 sbits; 531 532 redo: 533 hlist_for_each_entry_safe(pol, tmp, list, bydst) { 534 unsigned int h; 535 536 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits); 537 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr, 538 pol->family, nhashmask, dbits, sbits); 539 if (!entry0 || pol->xdo.type == XFRM_DEV_OFFLOAD_PACKET) { 540 hlist_del_rcu(&pol->bydst); 541 hlist_add_head_rcu(&pol->bydst, ndsttable + h); 542 h0 = h; 543 } else { 544 if (h != h0) 545 continue; 546 hlist_del_rcu(&pol->bydst); 547 hlist_add_behind_rcu(&pol->bydst, entry0); 548 } 549 entry0 = &pol->bydst; 550 } 551 if (!hlist_empty(list)) { 552 entry0 = NULL; 553 goto redo; 554 } 555 } 556 557 static void xfrm_idx_hash_transfer(struct hlist_head *list, 558 struct hlist_head *nidxtable, 559 unsigned int nhashmask) 560 { 561 struct hlist_node *tmp; 562 struct xfrm_policy *pol; 563 564 hlist_for_each_entry_safe(pol, tmp, list, byidx) { 565 unsigned int h; 566 567 h = __idx_hash(pol->index, nhashmask); 568 hlist_add_head(&pol->byidx, nidxtable+h); 569 } 570 } 571 572 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask) 573 { 574 return ((old_hmask + 1) << 1) - 1; 575 } 576 577 static void xfrm_bydst_resize(struct net *net, int dir) 578 { 579 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 580 unsigned int nhashmask = xfrm_new_hash_mask(hmask); 581 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head); 582 struct hlist_head *ndst = xfrm_hash_alloc(nsize); 583 struct hlist_head *odst; 584 int i; 585 586 if (!ndst) 587 return; 588 589 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 590 write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation); 591 592 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table, 593 lockdep_is_held(&net->xfrm.xfrm_policy_lock)); 594 595 for (i = hmask; i >= 0; i--) 596 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir); 597 598 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst); 599 net->xfrm.policy_bydst[dir].hmask = nhashmask; 600 601 write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation); 602 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 603 604 synchronize_rcu(); 605 606 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head)); 607 } 608 609 static void xfrm_byidx_resize(struct net *net) 610 { 611 unsigned int hmask = net->xfrm.policy_idx_hmask; 612 unsigned int nhashmask = xfrm_new_hash_mask(hmask); 613 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head); 614 struct hlist_head *oidx = net->xfrm.policy_byidx; 615 struct hlist_head *nidx = xfrm_hash_alloc(nsize); 616 int i; 617 618 if (!nidx) 619 return; 620 621 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 622 623 for (i = hmask; i >= 0; i--) 624 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask); 625 626 net->xfrm.policy_byidx = nidx; 627 net->xfrm.policy_idx_hmask = nhashmask; 628 629 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 630 631 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head)); 632 } 633 634 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total) 635 { 636 unsigned int cnt = net->xfrm.policy_count[dir]; 637 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 638 639 if (total) 640 *total += cnt; 641 642 if ((hmask + 1) < xfrm_policy_hashmax && 643 cnt > hmask) 644 return 1; 645 646 return 0; 647 } 648 649 static inline int xfrm_byidx_should_resize(struct net *net, int total) 650 { 651 unsigned int hmask = net->xfrm.policy_idx_hmask; 652 653 if ((hmask + 1) < xfrm_policy_hashmax && 654 total > hmask) 655 return 1; 656 657 return 0; 658 } 659 660 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si) 661 { 662 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN]; 663 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT]; 664 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD]; 665 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX]; 666 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX]; 667 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX]; 668 si->spdhcnt = net->xfrm.policy_idx_hmask; 669 si->spdhmcnt = xfrm_policy_hashmax; 670 } 671 EXPORT_SYMBOL(xfrm_spd_getinfo); 672 673 static DEFINE_MUTEX(hash_resize_mutex); 674 static void xfrm_hash_resize(struct work_struct *work) 675 { 676 struct net *net = container_of(work, struct net, xfrm.policy_hash_work); 677 int dir, total; 678 679 mutex_lock(&hash_resize_mutex); 680 681 total = 0; 682 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 683 if (xfrm_bydst_should_resize(net, dir, &total)) 684 xfrm_bydst_resize(net, dir); 685 } 686 if (xfrm_byidx_should_resize(net, total)) 687 xfrm_byidx_resize(net); 688 689 mutex_unlock(&hash_resize_mutex); 690 } 691 692 /* Make sure *pol can be inserted into fastbin. 693 * Useful to check that later insert requests will be successful 694 * (provided xfrm_policy_lock is held throughout). 695 */ 696 static struct xfrm_pol_inexact_bin * 697 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir) 698 { 699 struct xfrm_pol_inexact_bin *bin, *prev; 700 struct xfrm_pol_inexact_key k = { 701 .family = pol->family, 702 .type = pol->type, 703 .dir = dir, 704 .if_id = pol->if_id, 705 }; 706 struct net *net = xp_net(pol); 707 708 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 709 710 write_pnet(&k.net, net); 711 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k, 712 xfrm_pol_inexact_params); 713 if (bin) 714 return bin; 715 716 bin = kzalloc(sizeof(*bin), GFP_ATOMIC); 717 if (!bin) 718 return NULL; 719 720 bin->k = k; 721 INIT_HLIST_HEAD(&bin->hhead); 722 bin->root_d = RB_ROOT; 723 bin->root_s = RB_ROOT; 724 seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock); 725 726 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table, 727 &bin->k, &bin->head, 728 xfrm_pol_inexact_params); 729 if (!prev) { 730 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins); 731 return bin; 732 } 733 734 kfree(bin); 735 736 return IS_ERR(prev) ? NULL : prev; 737 } 738 739 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr, 740 int family, u8 prefixlen) 741 { 742 if (xfrm_addr_any(addr, family)) 743 return true; 744 745 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6) 746 return true; 747 748 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4) 749 return true; 750 751 return false; 752 } 753 754 static bool 755 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy) 756 { 757 const xfrm_address_t *addr; 758 bool saddr_any, daddr_any; 759 u8 prefixlen; 760 761 addr = &policy->selector.saddr; 762 prefixlen = policy->selector.prefixlen_s; 763 764 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr, 765 policy->family, 766 prefixlen); 767 addr = &policy->selector.daddr; 768 prefixlen = policy->selector.prefixlen_d; 769 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr, 770 policy->family, 771 prefixlen); 772 return saddr_any && daddr_any; 773 } 774 775 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node, 776 const xfrm_address_t *addr, u8 prefixlen) 777 { 778 node->addr = *addr; 779 node->prefixlen = prefixlen; 780 } 781 782 static struct xfrm_pol_inexact_node * 783 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen) 784 { 785 struct xfrm_pol_inexact_node *node; 786 787 node = kzalloc(sizeof(*node), GFP_ATOMIC); 788 if (node) 789 xfrm_pol_inexact_node_init(node, addr, prefixlen); 790 791 return node; 792 } 793 794 static int xfrm_policy_addr_delta(const xfrm_address_t *a, 795 const xfrm_address_t *b, 796 u8 prefixlen, u16 family) 797 { 798 u32 ma, mb, mask; 799 unsigned int pdw, pbi; 800 int delta = 0; 801 802 switch (family) { 803 case AF_INET: 804 if (prefixlen == 0) 805 return 0; 806 mask = ~0U << (32 - prefixlen); 807 ma = ntohl(a->a4) & mask; 808 mb = ntohl(b->a4) & mask; 809 if (ma < mb) 810 delta = -1; 811 else if (ma > mb) 812 delta = 1; 813 break; 814 case AF_INET6: 815 pdw = prefixlen >> 5; 816 pbi = prefixlen & 0x1f; 817 818 if (pdw) { 819 delta = memcmp(a->a6, b->a6, pdw << 2); 820 if (delta) 821 return delta; 822 } 823 if (pbi) { 824 mask = ~0U << (32 - pbi); 825 ma = ntohl(a->a6[pdw]) & mask; 826 mb = ntohl(b->a6[pdw]) & mask; 827 if (ma < mb) 828 delta = -1; 829 else if (ma > mb) 830 delta = 1; 831 } 832 break; 833 default: 834 break; 835 } 836 837 return delta; 838 } 839 840 static void xfrm_policy_inexact_list_reinsert(struct net *net, 841 struct xfrm_pol_inexact_node *n, 842 u16 family) 843 { 844 unsigned int matched_s, matched_d; 845 struct xfrm_policy *policy, *p; 846 847 matched_s = 0; 848 matched_d = 0; 849 850 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) { 851 struct hlist_node *newpos = NULL; 852 bool matches_s, matches_d; 853 854 if (policy->walk.dead || !policy->bydst_reinsert) 855 continue; 856 857 WARN_ON_ONCE(policy->family != family); 858 859 policy->bydst_reinsert = false; 860 hlist_for_each_entry(p, &n->hhead, bydst) { 861 if (policy->priority > p->priority) 862 newpos = &p->bydst; 863 else if (policy->priority == p->priority && 864 policy->pos > p->pos) 865 newpos = &p->bydst; 866 else 867 break; 868 } 869 870 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET) 871 hlist_add_behind_rcu(&policy->bydst, newpos); 872 else 873 hlist_add_head_rcu(&policy->bydst, &n->hhead); 874 875 /* paranoia checks follow. 876 * Check that the reinserted policy matches at least 877 * saddr or daddr for current node prefix. 878 * 879 * Matching both is fine, matching saddr in one policy 880 * (but not daddr) and then matching only daddr in another 881 * is a bug. 882 */ 883 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr, 884 &n->addr, 885 n->prefixlen, 886 family) == 0; 887 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr, 888 &n->addr, 889 n->prefixlen, 890 family) == 0; 891 if (matches_s && matches_d) 892 continue; 893 894 WARN_ON_ONCE(!matches_s && !matches_d); 895 if (matches_s) 896 matched_s++; 897 if (matches_d) 898 matched_d++; 899 WARN_ON_ONCE(matched_s && matched_d); 900 } 901 } 902 903 static void xfrm_policy_inexact_node_reinsert(struct net *net, 904 struct xfrm_pol_inexact_node *n, 905 struct rb_root *new, 906 u16 family) 907 { 908 struct xfrm_pol_inexact_node *node; 909 struct rb_node **p, *parent; 910 911 /* we should not have another subtree here */ 912 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root)); 913 restart: 914 parent = NULL; 915 p = &new->rb_node; 916 while (*p) { 917 u8 prefixlen; 918 int delta; 919 920 parent = *p; 921 node = rb_entry(*p, struct xfrm_pol_inexact_node, node); 922 923 prefixlen = min(node->prefixlen, n->prefixlen); 924 925 delta = xfrm_policy_addr_delta(&n->addr, &node->addr, 926 prefixlen, family); 927 if (delta < 0) { 928 p = &parent->rb_left; 929 } else if (delta > 0) { 930 p = &parent->rb_right; 931 } else { 932 bool same_prefixlen = node->prefixlen == n->prefixlen; 933 struct xfrm_policy *tmp; 934 935 hlist_for_each_entry(tmp, &n->hhead, bydst) { 936 tmp->bydst_reinsert = true; 937 hlist_del_rcu(&tmp->bydst); 938 } 939 940 node->prefixlen = prefixlen; 941 942 xfrm_policy_inexact_list_reinsert(net, node, family); 943 944 if (same_prefixlen) { 945 kfree_rcu(n, rcu); 946 return; 947 } 948 949 rb_erase(*p, new); 950 kfree_rcu(n, rcu); 951 n = node; 952 goto restart; 953 } 954 } 955 956 rb_link_node_rcu(&n->node, parent, p); 957 rb_insert_color(&n->node, new); 958 } 959 960 /* merge nodes v and n */ 961 static void xfrm_policy_inexact_node_merge(struct net *net, 962 struct xfrm_pol_inexact_node *v, 963 struct xfrm_pol_inexact_node *n, 964 u16 family) 965 { 966 struct xfrm_pol_inexact_node *node; 967 struct xfrm_policy *tmp; 968 struct rb_node *rnode; 969 970 /* To-be-merged node v has a subtree. 971 * 972 * Dismantle it and insert its nodes to n->root. 973 */ 974 while ((rnode = rb_first(&v->root)) != NULL) { 975 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node); 976 rb_erase(&node->node, &v->root); 977 xfrm_policy_inexact_node_reinsert(net, node, &n->root, 978 family); 979 } 980 981 hlist_for_each_entry(tmp, &v->hhead, bydst) { 982 tmp->bydst_reinsert = true; 983 hlist_del_rcu(&tmp->bydst); 984 } 985 986 xfrm_policy_inexact_list_reinsert(net, n, family); 987 } 988 989 static struct xfrm_pol_inexact_node * 990 xfrm_policy_inexact_insert_node(struct net *net, 991 struct rb_root *root, 992 xfrm_address_t *addr, 993 u16 family, u8 prefixlen, u8 dir) 994 { 995 struct xfrm_pol_inexact_node *cached = NULL; 996 struct rb_node **p, *parent = NULL; 997 struct xfrm_pol_inexact_node *node; 998 999 p = &root->rb_node; 1000 while (*p) { 1001 int delta; 1002 1003 parent = *p; 1004 node = rb_entry(*p, struct xfrm_pol_inexact_node, node); 1005 1006 delta = xfrm_policy_addr_delta(addr, &node->addr, 1007 node->prefixlen, 1008 family); 1009 if (delta == 0 && prefixlen >= node->prefixlen) { 1010 WARN_ON_ONCE(cached); /* ipsec policies got lost */ 1011 return node; 1012 } 1013 1014 if (delta < 0) 1015 p = &parent->rb_left; 1016 else 1017 p = &parent->rb_right; 1018 1019 if (prefixlen < node->prefixlen) { 1020 delta = xfrm_policy_addr_delta(addr, &node->addr, 1021 prefixlen, 1022 family); 1023 if (delta) 1024 continue; 1025 1026 /* This node is a subnet of the new prefix. It needs 1027 * to be removed and re-inserted with the smaller 1028 * prefix and all nodes that are now also covered 1029 * by the reduced prefixlen. 1030 */ 1031 rb_erase(&node->node, root); 1032 1033 if (!cached) { 1034 xfrm_pol_inexact_node_init(node, addr, 1035 prefixlen); 1036 cached = node; 1037 } else { 1038 /* This node also falls within the new 1039 * prefixlen. Merge the to-be-reinserted 1040 * node and this one. 1041 */ 1042 xfrm_policy_inexact_node_merge(net, node, 1043 cached, family); 1044 kfree_rcu(node, rcu); 1045 } 1046 1047 /* restart */ 1048 p = &root->rb_node; 1049 parent = NULL; 1050 } 1051 } 1052 1053 node = cached; 1054 if (!node) { 1055 node = xfrm_pol_inexact_node_alloc(addr, prefixlen); 1056 if (!node) 1057 return NULL; 1058 } 1059 1060 rb_link_node_rcu(&node->node, parent, p); 1061 rb_insert_color(&node->node, root); 1062 1063 return node; 1064 } 1065 1066 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm) 1067 { 1068 struct xfrm_pol_inexact_node *node; 1069 struct rb_node *rn = rb_first(r); 1070 1071 while (rn) { 1072 node = rb_entry(rn, struct xfrm_pol_inexact_node, node); 1073 1074 xfrm_policy_inexact_gc_tree(&node->root, rm); 1075 rn = rb_next(rn); 1076 1077 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) { 1078 WARN_ON_ONCE(rm); 1079 continue; 1080 } 1081 1082 rb_erase(&node->node, r); 1083 kfree_rcu(node, rcu); 1084 } 1085 } 1086 1087 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit) 1088 { 1089 write_seqcount_begin(&b->count); 1090 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit); 1091 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit); 1092 write_seqcount_end(&b->count); 1093 1094 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) || 1095 !hlist_empty(&b->hhead)) { 1096 WARN_ON_ONCE(net_exit); 1097 return; 1098 } 1099 1100 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head, 1101 xfrm_pol_inexact_params) == 0) { 1102 list_del(&b->inexact_bins); 1103 kfree_rcu(b, rcu); 1104 } 1105 } 1106 1107 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b) 1108 { 1109 struct net *net = read_pnet(&b->k.net); 1110 1111 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1112 __xfrm_policy_inexact_prune_bin(b, false); 1113 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1114 } 1115 1116 static void __xfrm_policy_inexact_flush(struct net *net) 1117 { 1118 struct xfrm_pol_inexact_bin *bin, *t; 1119 1120 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1121 1122 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins) 1123 __xfrm_policy_inexact_prune_bin(bin, false); 1124 } 1125 1126 static struct hlist_head * 1127 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin, 1128 struct xfrm_policy *policy, u8 dir) 1129 { 1130 struct xfrm_pol_inexact_node *n; 1131 struct net *net; 1132 1133 net = xp_net(policy); 1134 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1135 1136 if (xfrm_policy_inexact_insert_use_any_list(policy)) 1137 return &bin->hhead; 1138 1139 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr, 1140 policy->family, 1141 policy->selector.prefixlen_d)) { 1142 write_seqcount_begin(&bin->count); 1143 n = xfrm_policy_inexact_insert_node(net, 1144 &bin->root_s, 1145 &policy->selector.saddr, 1146 policy->family, 1147 policy->selector.prefixlen_s, 1148 dir); 1149 write_seqcount_end(&bin->count); 1150 if (!n) 1151 return NULL; 1152 1153 return &n->hhead; 1154 } 1155 1156 /* daddr is fixed */ 1157 write_seqcount_begin(&bin->count); 1158 n = xfrm_policy_inexact_insert_node(net, 1159 &bin->root_d, 1160 &policy->selector.daddr, 1161 policy->family, 1162 policy->selector.prefixlen_d, dir); 1163 write_seqcount_end(&bin->count); 1164 if (!n) 1165 return NULL; 1166 1167 /* saddr is wildcard */ 1168 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr, 1169 policy->family, 1170 policy->selector.prefixlen_s)) 1171 return &n->hhead; 1172 1173 write_seqcount_begin(&bin->count); 1174 n = xfrm_policy_inexact_insert_node(net, 1175 &n->root, 1176 &policy->selector.saddr, 1177 policy->family, 1178 policy->selector.prefixlen_s, dir); 1179 write_seqcount_end(&bin->count); 1180 if (!n) 1181 return NULL; 1182 1183 return &n->hhead; 1184 } 1185 1186 static struct xfrm_policy * 1187 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl) 1188 { 1189 struct xfrm_pol_inexact_bin *bin; 1190 struct xfrm_policy *delpol; 1191 struct hlist_head *chain; 1192 struct net *net; 1193 1194 bin = xfrm_policy_inexact_alloc_bin(policy, dir); 1195 if (!bin) 1196 return ERR_PTR(-ENOMEM); 1197 1198 net = xp_net(policy); 1199 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1200 1201 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir); 1202 if (!chain) { 1203 __xfrm_policy_inexact_prune_bin(bin, false); 1204 return ERR_PTR(-ENOMEM); 1205 } 1206 1207 delpol = xfrm_policy_insert_list(chain, policy, excl); 1208 if (delpol && excl) { 1209 __xfrm_policy_inexact_prune_bin(bin, false); 1210 return ERR_PTR(-EEXIST); 1211 } 1212 1213 chain = &net->xfrm.policy_inexact[dir]; 1214 xfrm_policy_insert_inexact_list(chain, policy); 1215 1216 if (delpol) 1217 __xfrm_policy_inexact_prune_bin(bin, false); 1218 1219 return delpol; 1220 } 1221 1222 static void xfrm_hash_rebuild(struct work_struct *work) 1223 { 1224 struct net *net = container_of(work, struct net, 1225 xfrm.policy_hthresh.work); 1226 unsigned int hmask; 1227 struct xfrm_policy *pol; 1228 struct xfrm_policy *policy; 1229 struct hlist_head *chain; 1230 struct hlist_head *odst; 1231 struct hlist_node *newpos; 1232 int i; 1233 int dir; 1234 unsigned seq; 1235 u8 lbits4, rbits4, lbits6, rbits6; 1236 1237 mutex_lock(&hash_resize_mutex); 1238 1239 /* read selector prefixlen thresholds */ 1240 do { 1241 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock); 1242 1243 lbits4 = net->xfrm.policy_hthresh.lbits4; 1244 rbits4 = net->xfrm.policy_hthresh.rbits4; 1245 lbits6 = net->xfrm.policy_hthresh.lbits6; 1246 rbits6 = net->xfrm.policy_hthresh.rbits6; 1247 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq)); 1248 1249 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1250 write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation); 1251 1252 /* make sure that we can insert the indirect policies again before 1253 * we start with destructive action. 1254 */ 1255 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) { 1256 struct xfrm_pol_inexact_bin *bin; 1257 u8 dbits, sbits; 1258 1259 if (policy->walk.dead) 1260 continue; 1261 1262 dir = xfrm_policy_id2dir(policy->index); 1263 if (dir >= XFRM_POLICY_MAX) 1264 continue; 1265 1266 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) { 1267 if (policy->family == AF_INET) { 1268 dbits = rbits4; 1269 sbits = lbits4; 1270 } else { 1271 dbits = rbits6; 1272 sbits = lbits6; 1273 } 1274 } else { 1275 if (policy->family == AF_INET) { 1276 dbits = lbits4; 1277 sbits = rbits4; 1278 } else { 1279 dbits = lbits6; 1280 sbits = rbits6; 1281 } 1282 } 1283 1284 if (policy->selector.prefixlen_d < dbits || 1285 policy->selector.prefixlen_s < sbits) 1286 continue; 1287 1288 bin = xfrm_policy_inexact_alloc_bin(policy, dir); 1289 if (!bin) 1290 goto out_unlock; 1291 1292 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir)) 1293 goto out_unlock; 1294 } 1295 1296 /* reset the bydst and inexact table in all directions */ 1297 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 1298 struct hlist_node *n; 1299 1300 hlist_for_each_entry_safe(policy, n, 1301 &net->xfrm.policy_inexact[dir], 1302 bydst_inexact_list) { 1303 hlist_del_rcu(&policy->bydst); 1304 hlist_del_init(&policy->bydst_inexact_list); 1305 } 1306 1307 hmask = net->xfrm.policy_bydst[dir].hmask; 1308 odst = net->xfrm.policy_bydst[dir].table; 1309 for (i = hmask; i >= 0; i--) { 1310 hlist_for_each_entry_safe(policy, n, odst + i, bydst) 1311 hlist_del_rcu(&policy->bydst); 1312 } 1313 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) { 1314 /* dir out => dst = remote, src = local */ 1315 net->xfrm.policy_bydst[dir].dbits4 = rbits4; 1316 net->xfrm.policy_bydst[dir].sbits4 = lbits4; 1317 net->xfrm.policy_bydst[dir].dbits6 = rbits6; 1318 net->xfrm.policy_bydst[dir].sbits6 = lbits6; 1319 } else { 1320 /* dir in/fwd => dst = local, src = remote */ 1321 net->xfrm.policy_bydst[dir].dbits4 = lbits4; 1322 net->xfrm.policy_bydst[dir].sbits4 = rbits4; 1323 net->xfrm.policy_bydst[dir].dbits6 = lbits6; 1324 net->xfrm.policy_bydst[dir].sbits6 = rbits6; 1325 } 1326 } 1327 1328 /* re-insert all policies by order of creation */ 1329 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) { 1330 if (policy->walk.dead) 1331 continue; 1332 dir = xfrm_policy_id2dir(policy->index); 1333 if (dir >= XFRM_POLICY_MAX) { 1334 /* skip socket policies */ 1335 continue; 1336 } 1337 newpos = NULL; 1338 chain = policy_hash_bysel(net, &policy->selector, 1339 policy->family, dir); 1340 1341 if (!chain) { 1342 void *p = xfrm_policy_inexact_insert(policy, dir, 0); 1343 1344 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p)); 1345 continue; 1346 } 1347 1348 hlist_for_each_entry(pol, chain, bydst) { 1349 if (policy->priority >= pol->priority) 1350 newpos = &pol->bydst; 1351 else 1352 break; 1353 } 1354 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET) 1355 hlist_add_behind_rcu(&policy->bydst, newpos); 1356 else 1357 hlist_add_head_rcu(&policy->bydst, chain); 1358 } 1359 1360 out_unlock: 1361 __xfrm_policy_inexact_flush(net); 1362 write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation); 1363 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1364 1365 mutex_unlock(&hash_resize_mutex); 1366 } 1367 1368 void xfrm_policy_hash_rebuild(struct net *net) 1369 { 1370 schedule_work(&net->xfrm.policy_hthresh.work); 1371 } 1372 EXPORT_SYMBOL(xfrm_policy_hash_rebuild); 1373 1374 /* Generate new index... KAME seems to generate them ordered by cost 1375 * of an absolute inpredictability of ordering of rules. This will not pass. */ 1376 static u32 xfrm_gen_index(struct net *net, int dir, u32 index) 1377 { 1378 for (;;) { 1379 struct hlist_head *list; 1380 struct xfrm_policy *p; 1381 u32 idx; 1382 int found; 1383 1384 if (!index) { 1385 idx = (net->xfrm.idx_generator | dir); 1386 net->xfrm.idx_generator += 8; 1387 } else { 1388 idx = index; 1389 index = 0; 1390 } 1391 1392 if (idx == 0) 1393 idx = 8; 1394 list = net->xfrm.policy_byidx + idx_hash(net, idx); 1395 found = 0; 1396 hlist_for_each_entry(p, list, byidx) { 1397 if (p->index == idx) { 1398 found = 1; 1399 break; 1400 } 1401 } 1402 if (!found) 1403 return idx; 1404 } 1405 } 1406 1407 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2) 1408 { 1409 u32 *p1 = (u32 *) s1; 1410 u32 *p2 = (u32 *) s2; 1411 int len = sizeof(struct xfrm_selector) / sizeof(u32); 1412 int i; 1413 1414 for (i = 0; i < len; i++) { 1415 if (p1[i] != p2[i]) 1416 return 1; 1417 } 1418 1419 return 0; 1420 } 1421 1422 static void xfrm_policy_requeue(struct xfrm_policy *old, 1423 struct xfrm_policy *new) 1424 { 1425 struct xfrm_policy_queue *pq = &old->polq; 1426 struct sk_buff_head list; 1427 1428 if (skb_queue_empty(&pq->hold_queue)) 1429 return; 1430 1431 __skb_queue_head_init(&list); 1432 1433 spin_lock_bh(&pq->hold_queue.lock); 1434 skb_queue_splice_init(&pq->hold_queue, &list); 1435 if (del_timer(&pq->hold_timer)) 1436 xfrm_pol_put(old); 1437 spin_unlock_bh(&pq->hold_queue.lock); 1438 1439 pq = &new->polq; 1440 1441 spin_lock_bh(&pq->hold_queue.lock); 1442 skb_queue_splice(&list, &pq->hold_queue); 1443 pq->timeout = XFRM_QUEUE_TMO_MIN; 1444 if (!mod_timer(&pq->hold_timer, jiffies)) 1445 xfrm_pol_hold(new); 1446 spin_unlock_bh(&pq->hold_queue.lock); 1447 } 1448 1449 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark, 1450 struct xfrm_policy *pol) 1451 { 1452 return mark->v == pol->mark.v && mark->m == pol->mark.m; 1453 } 1454 1455 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed) 1456 { 1457 const struct xfrm_pol_inexact_key *k = data; 1458 u32 a = k->type << 24 | k->dir << 16 | k->family; 1459 1460 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)), 1461 seed); 1462 } 1463 1464 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed) 1465 { 1466 const struct xfrm_pol_inexact_bin *b = data; 1467 1468 return xfrm_pol_bin_key(&b->k, 0, seed); 1469 } 1470 1471 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg, 1472 const void *ptr) 1473 { 1474 const struct xfrm_pol_inexact_key *key = arg->key; 1475 const struct xfrm_pol_inexact_bin *b = ptr; 1476 int ret; 1477 1478 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net))) 1479 return -1; 1480 1481 ret = b->k.dir ^ key->dir; 1482 if (ret) 1483 return ret; 1484 1485 ret = b->k.type ^ key->type; 1486 if (ret) 1487 return ret; 1488 1489 ret = b->k.family ^ key->family; 1490 if (ret) 1491 return ret; 1492 1493 return b->k.if_id ^ key->if_id; 1494 } 1495 1496 static const struct rhashtable_params xfrm_pol_inexact_params = { 1497 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head), 1498 .hashfn = xfrm_pol_bin_key, 1499 .obj_hashfn = xfrm_pol_bin_obj, 1500 .obj_cmpfn = xfrm_pol_bin_cmp, 1501 .automatic_shrinking = true, 1502 }; 1503 1504 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain, 1505 struct xfrm_policy *policy) 1506 { 1507 struct xfrm_policy *pol, *delpol = NULL; 1508 struct hlist_node *newpos = NULL; 1509 int i = 0; 1510 1511 hlist_for_each_entry(pol, chain, bydst_inexact_list) { 1512 if (pol->type == policy->type && 1513 pol->if_id == policy->if_id && 1514 !selector_cmp(&pol->selector, &policy->selector) && 1515 xfrm_policy_mark_match(&policy->mark, pol) && 1516 xfrm_sec_ctx_match(pol->security, policy->security) && 1517 !WARN_ON(delpol)) { 1518 delpol = pol; 1519 if (policy->priority > pol->priority) 1520 continue; 1521 } else if (policy->priority >= pol->priority) { 1522 newpos = &pol->bydst_inexact_list; 1523 continue; 1524 } 1525 if (delpol) 1526 break; 1527 } 1528 1529 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET) 1530 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos); 1531 else 1532 hlist_add_head_rcu(&policy->bydst_inexact_list, chain); 1533 1534 hlist_for_each_entry(pol, chain, bydst_inexact_list) { 1535 pol->pos = i; 1536 i++; 1537 } 1538 } 1539 1540 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain, 1541 struct xfrm_policy *policy, 1542 bool excl) 1543 { 1544 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL; 1545 1546 hlist_for_each_entry(pol, chain, bydst) { 1547 if (pol->type == policy->type && 1548 pol->if_id == policy->if_id && 1549 !selector_cmp(&pol->selector, &policy->selector) && 1550 xfrm_policy_mark_match(&policy->mark, pol) && 1551 xfrm_sec_ctx_match(pol->security, policy->security) && 1552 !WARN_ON(delpol)) { 1553 if (excl) 1554 return ERR_PTR(-EEXIST); 1555 delpol = pol; 1556 if (policy->priority > pol->priority) 1557 continue; 1558 } else if (policy->priority >= pol->priority) { 1559 newpos = pol; 1560 continue; 1561 } 1562 if (delpol) 1563 break; 1564 } 1565 1566 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET) 1567 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst); 1568 else 1569 /* Packet offload policies enter to the head 1570 * to speed-up lookups. 1571 */ 1572 hlist_add_head_rcu(&policy->bydst, chain); 1573 1574 return delpol; 1575 } 1576 1577 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl) 1578 { 1579 struct net *net = xp_net(policy); 1580 struct xfrm_policy *delpol; 1581 struct hlist_head *chain; 1582 1583 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1584 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir); 1585 if (chain) 1586 delpol = xfrm_policy_insert_list(chain, policy, excl); 1587 else 1588 delpol = xfrm_policy_inexact_insert(policy, dir, excl); 1589 1590 if (IS_ERR(delpol)) { 1591 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1592 return PTR_ERR(delpol); 1593 } 1594 1595 __xfrm_policy_link(policy, dir); 1596 1597 /* After previous checking, family can either be AF_INET or AF_INET6 */ 1598 if (policy->family == AF_INET) 1599 rt_genid_bump_ipv4(net); 1600 else 1601 rt_genid_bump_ipv6(net); 1602 1603 if (delpol) { 1604 xfrm_policy_requeue(delpol, policy); 1605 __xfrm_policy_unlink(delpol, dir); 1606 } 1607 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index); 1608 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index)); 1609 policy->curlft.add_time = ktime_get_real_seconds(); 1610 policy->curlft.use_time = 0; 1611 if (!mod_timer(&policy->timer, jiffies + HZ)) 1612 xfrm_pol_hold(policy); 1613 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1614 1615 if (delpol) 1616 xfrm_policy_kill(delpol); 1617 else if (xfrm_bydst_should_resize(net, dir, NULL)) 1618 schedule_work(&net->xfrm.policy_hash_work); 1619 1620 return 0; 1621 } 1622 EXPORT_SYMBOL(xfrm_policy_insert); 1623 1624 static struct xfrm_policy * 1625 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark, 1626 u32 if_id, u8 type, int dir, struct xfrm_selector *sel, 1627 struct xfrm_sec_ctx *ctx) 1628 { 1629 struct xfrm_policy *pol; 1630 1631 if (!chain) 1632 return NULL; 1633 1634 hlist_for_each_entry(pol, chain, bydst) { 1635 if (pol->type == type && 1636 pol->if_id == if_id && 1637 xfrm_policy_mark_match(mark, pol) && 1638 !selector_cmp(sel, &pol->selector) && 1639 xfrm_sec_ctx_match(ctx, pol->security)) 1640 return pol; 1641 } 1642 1643 return NULL; 1644 } 1645 1646 struct xfrm_policy * 1647 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id, 1648 u8 type, int dir, struct xfrm_selector *sel, 1649 struct xfrm_sec_ctx *ctx, int delete, int *err) 1650 { 1651 struct xfrm_pol_inexact_bin *bin = NULL; 1652 struct xfrm_policy *pol, *ret = NULL; 1653 struct hlist_head *chain; 1654 1655 *err = 0; 1656 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1657 chain = policy_hash_bysel(net, sel, sel->family, dir); 1658 if (!chain) { 1659 struct xfrm_pol_inexact_candidates cand; 1660 int i; 1661 1662 bin = xfrm_policy_inexact_lookup(net, type, 1663 sel->family, dir, if_id); 1664 if (!bin) { 1665 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1666 return NULL; 1667 } 1668 1669 if (!xfrm_policy_find_inexact_candidates(&cand, bin, 1670 &sel->saddr, 1671 &sel->daddr)) { 1672 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1673 return NULL; 1674 } 1675 1676 pol = NULL; 1677 for (i = 0; i < ARRAY_SIZE(cand.res); i++) { 1678 struct xfrm_policy *tmp; 1679 1680 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark, 1681 if_id, type, dir, 1682 sel, ctx); 1683 if (!tmp) 1684 continue; 1685 1686 if (!pol || tmp->pos < pol->pos) 1687 pol = tmp; 1688 } 1689 } else { 1690 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir, 1691 sel, ctx); 1692 } 1693 1694 if (pol) { 1695 xfrm_pol_hold(pol); 1696 if (delete) { 1697 *err = security_xfrm_policy_delete(pol->security); 1698 if (*err) { 1699 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1700 return pol; 1701 } 1702 __xfrm_policy_unlink(pol, dir); 1703 } 1704 ret = pol; 1705 } 1706 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1707 1708 if (ret && delete) 1709 xfrm_policy_kill(ret); 1710 if (bin && delete) 1711 xfrm_policy_inexact_prune_bin(bin); 1712 return ret; 1713 } 1714 EXPORT_SYMBOL(xfrm_policy_bysel_ctx); 1715 1716 struct xfrm_policy * 1717 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id, 1718 u8 type, int dir, u32 id, int delete, int *err) 1719 { 1720 struct xfrm_policy *pol, *ret; 1721 struct hlist_head *chain; 1722 1723 *err = -ENOENT; 1724 if (xfrm_policy_id2dir(id) != dir) 1725 return NULL; 1726 1727 *err = 0; 1728 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1729 chain = net->xfrm.policy_byidx + idx_hash(net, id); 1730 ret = NULL; 1731 hlist_for_each_entry(pol, chain, byidx) { 1732 if (pol->type == type && pol->index == id && 1733 pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) { 1734 xfrm_pol_hold(pol); 1735 if (delete) { 1736 *err = security_xfrm_policy_delete( 1737 pol->security); 1738 if (*err) { 1739 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1740 return pol; 1741 } 1742 __xfrm_policy_unlink(pol, dir); 1743 } 1744 ret = pol; 1745 break; 1746 } 1747 } 1748 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1749 1750 if (ret && delete) 1751 xfrm_policy_kill(ret); 1752 return ret; 1753 } 1754 EXPORT_SYMBOL(xfrm_policy_byid); 1755 1756 #ifdef CONFIG_SECURITY_NETWORK_XFRM 1757 static inline int 1758 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid) 1759 { 1760 struct xfrm_policy *pol; 1761 int err = 0; 1762 1763 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1764 if (pol->walk.dead || 1765 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX || 1766 pol->type != type) 1767 continue; 1768 1769 err = security_xfrm_policy_delete(pol->security); 1770 if (err) { 1771 xfrm_audit_policy_delete(pol, 0, task_valid); 1772 return err; 1773 } 1774 } 1775 return err; 1776 } 1777 1778 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net, 1779 struct net_device *dev, 1780 bool task_valid) 1781 { 1782 struct xfrm_policy *pol; 1783 int err = 0; 1784 1785 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1786 if (pol->walk.dead || 1787 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX || 1788 pol->xdo.dev != dev) 1789 continue; 1790 1791 err = security_xfrm_policy_delete(pol->security); 1792 if (err) { 1793 xfrm_audit_policy_delete(pol, 0, task_valid); 1794 return err; 1795 } 1796 } 1797 return err; 1798 } 1799 #else 1800 static inline int 1801 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid) 1802 { 1803 return 0; 1804 } 1805 1806 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net, 1807 struct net_device *dev, 1808 bool task_valid) 1809 { 1810 return 0; 1811 } 1812 #endif 1813 1814 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid) 1815 { 1816 int dir, err = 0, cnt = 0; 1817 struct xfrm_policy *pol; 1818 1819 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1820 1821 err = xfrm_policy_flush_secctx_check(net, type, task_valid); 1822 if (err) 1823 goto out; 1824 1825 again: 1826 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1827 if (pol->walk.dead) 1828 continue; 1829 1830 dir = xfrm_policy_id2dir(pol->index); 1831 if (dir >= XFRM_POLICY_MAX || 1832 pol->type != type) 1833 continue; 1834 1835 __xfrm_policy_unlink(pol, dir); 1836 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1837 xfrm_dev_policy_delete(pol); 1838 cnt++; 1839 xfrm_audit_policy_delete(pol, 1, task_valid); 1840 xfrm_policy_kill(pol); 1841 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1842 goto again; 1843 } 1844 if (cnt) 1845 __xfrm_policy_inexact_flush(net); 1846 else 1847 err = -ESRCH; 1848 out: 1849 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1850 return err; 1851 } 1852 EXPORT_SYMBOL(xfrm_policy_flush); 1853 1854 int xfrm_dev_policy_flush(struct net *net, struct net_device *dev, 1855 bool task_valid) 1856 { 1857 int dir, err = 0, cnt = 0; 1858 struct xfrm_policy *pol; 1859 1860 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1861 1862 err = xfrm_dev_policy_flush_secctx_check(net, dev, task_valid); 1863 if (err) 1864 goto out; 1865 1866 again: 1867 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1868 if (pol->walk.dead) 1869 continue; 1870 1871 dir = xfrm_policy_id2dir(pol->index); 1872 if (dir >= XFRM_POLICY_MAX || 1873 pol->xdo.dev != dev) 1874 continue; 1875 1876 __xfrm_policy_unlink(pol, dir); 1877 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1878 xfrm_dev_policy_delete(pol); 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_dev_policy_delete(pol); 2330 xfrm_policy_kill(pol); 2331 return 0; 2332 } 2333 return -ENOENT; 2334 } 2335 EXPORT_SYMBOL(xfrm_policy_delete); 2336 2337 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol) 2338 { 2339 struct net *net = sock_net(sk); 2340 struct xfrm_policy *old_pol; 2341 2342 #ifdef CONFIG_XFRM_SUB_POLICY 2343 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN) 2344 return -EINVAL; 2345 #endif 2346 2347 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 2348 old_pol = rcu_dereference_protected(sk->sk_policy[dir], 2349 lockdep_is_held(&net->xfrm.xfrm_policy_lock)); 2350 if (pol) { 2351 pol->curlft.add_time = ktime_get_real_seconds(); 2352 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0); 2353 xfrm_sk_policy_link(pol, dir); 2354 } 2355 rcu_assign_pointer(sk->sk_policy[dir], pol); 2356 if (old_pol) { 2357 if (pol) 2358 xfrm_policy_requeue(old_pol, pol); 2359 2360 /* Unlinking succeeds always. This is the only function 2361 * allowed to delete or replace socket policy. 2362 */ 2363 xfrm_sk_policy_unlink(old_pol, dir); 2364 } 2365 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 2366 2367 if (old_pol) { 2368 xfrm_policy_kill(old_pol); 2369 } 2370 return 0; 2371 } 2372 2373 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir) 2374 { 2375 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC); 2376 struct net *net = xp_net(old); 2377 2378 if (newp) { 2379 newp->selector = old->selector; 2380 if (security_xfrm_policy_clone(old->security, 2381 &newp->security)) { 2382 kfree(newp); 2383 return NULL; /* ENOMEM */ 2384 } 2385 newp->lft = old->lft; 2386 newp->curlft = old->curlft; 2387 newp->mark = old->mark; 2388 newp->if_id = old->if_id; 2389 newp->action = old->action; 2390 newp->flags = old->flags; 2391 newp->xfrm_nr = old->xfrm_nr; 2392 newp->index = old->index; 2393 newp->type = old->type; 2394 newp->family = old->family; 2395 memcpy(newp->xfrm_vec, old->xfrm_vec, 2396 newp->xfrm_nr*sizeof(struct xfrm_tmpl)); 2397 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 2398 xfrm_sk_policy_link(newp, dir); 2399 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 2400 xfrm_pol_put(newp); 2401 } 2402 return newp; 2403 } 2404 2405 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk) 2406 { 2407 const struct xfrm_policy *p; 2408 struct xfrm_policy *np; 2409 int i, ret = 0; 2410 2411 rcu_read_lock(); 2412 for (i = 0; i < 2; i++) { 2413 p = rcu_dereference(osk->sk_policy[i]); 2414 if (p) { 2415 np = clone_policy(p, i); 2416 if (unlikely(!np)) { 2417 ret = -ENOMEM; 2418 break; 2419 } 2420 rcu_assign_pointer(sk->sk_policy[i], np); 2421 } 2422 } 2423 rcu_read_unlock(); 2424 return ret; 2425 } 2426 2427 static int 2428 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local, 2429 xfrm_address_t *remote, unsigned short family, u32 mark) 2430 { 2431 int err; 2432 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 2433 2434 if (unlikely(afinfo == NULL)) 2435 return -EINVAL; 2436 err = afinfo->get_saddr(net, oif, local, remote, mark); 2437 rcu_read_unlock(); 2438 return err; 2439 } 2440 2441 /* Resolve list of templates for the flow, given policy. */ 2442 2443 static int 2444 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl, 2445 struct xfrm_state **xfrm, unsigned short family) 2446 { 2447 struct net *net = xp_net(policy); 2448 int nx; 2449 int i, error; 2450 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family); 2451 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family); 2452 xfrm_address_t tmp; 2453 2454 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) { 2455 struct xfrm_state *x; 2456 xfrm_address_t *remote = daddr; 2457 xfrm_address_t *local = saddr; 2458 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i]; 2459 2460 if (tmpl->mode == XFRM_MODE_TUNNEL || 2461 tmpl->mode == XFRM_MODE_BEET) { 2462 remote = &tmpl->id.daddr; 2463 local = &tmpl->saddr; 2464 if (xfrm_addr_any(local, tmpl->encap_family)) { 2465 error = xfrm_get_saddr(net, fl->flowi_oif, 2466 &tmp, remote, 2467 tmpl->encap_family, 0); 2468 if (error) 2469 goto fail; 2470 local = &tmp; 2471 } 2472 } 2473 2474 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, 2475 family, policy->if_id); 2476 2477 if (x && x->km.state == XFRM_STATE_VALID) { 2478 xfrm[nx++] = x; 2479 daddr = remote; 2480 saddr = local; 2481 continue; 2482 } 2483 if (x) { 2484 error = (x->km.state == XFRM_STATE_ERROR ? 2485 -EINVAL : -EAGAIN); 2486 xfrm_state_put(x); 2487 } else if (error == -ESRCH) { 2488 error = -EAGAIN; 2489 } 2490 2491 if (!tmpl->optional) 2492 goto fail; 2493 } 2494 return nx; 2495 2496 fail: 2497 for (nx--; nx >= 0; nx--) 2498 xfrm_state_put(xfrm[nx]); 2499 return error; 2500 } 2501 2502 static int 2503 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl, 2504 struct xfrm_state **xfrm, unsigned short family) 2505 { 2506 struct xfrm_state *tp[XFRM_MAX_DEPTH]; 2507 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm; 2508 int cnx = 0; 2509 int error; 2510 int ret; 2511 int i; 2512 2513 for (i = 0; i < npols; i++) { 2514 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) { 2515 error = -ENOBUFS; 2516 goto fail; 2517 } 2518 2519 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family); 2520 if (ret < 0) { 2521 error = ret; 2522 goto fail; 2523 } else 2524 cnx += ret; 2525 } 2526 2527 /* found states are sorted for outbound processing */ 2528 if (npols > 1) 2529 xfrm_state_sort(xfrm, tpp, cnx, family); 2530 2531 return cnx; 2532 2533 fail: 2534 for (cnx--; cnx >= 0; cnx--) 2535 xfrm_state_put(tpp[cnx]); 2536 return error; 2537 2538 } 2539 2540 static int xfrm_get_tos(const struct flowi *fl, int family) 2541 { 2542 if (family == AF_INET) 2543 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos; 2544 2545 return 0; 2546 } 2547 2548 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family) 2549 { 2550 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 2551 struct dst_ops *dst_ops; 2552 struct xfrm_dst *xdst; 2553 2554 if (!afinfo) 2555 return ERR_PTR(-EINVAL); 2556 2557 switch (family) { 2558 case AF_INET: 2559 dst_ops = &net->xfrm.xfrm4_dst_ops; 2560 break; 2561 #if IS_ENABLED(CONFIG_IPV6) 2562 case AF_INET6: 2563 dst_ops = &net->xfrm.xfrm6_dst_ops; 2564 break; 2565 #endif 2566 default: 2567 BUG(); 2568 } 2569 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0); 2570 2571 if (likely(xdst)) { 2572 memset_after(xdst, 0, u.dst); 2573 } else 2574 xdst = ERR_PTR(-ENOBUFS); 2575 2576 rcu_read_unlock(); 2577 2578 return xdst; 2579 } 2580 2581 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst, 2582 int nfheader_len) 2583 { 2584 if (dst->ops->family == AF_INET6) { 2585 struct rt6_info *rt = (struct rt6_info *)dst; 2586 path->path_cookie = rt6_get_cookie(rt); 2587 path->u.rt6.rt6i_nfheader_len = nfheader_len; 2588 } 2589 } 2590 2591 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev, 2592 const struct flowi *fl) 2593 { 2594 const struct xfrm_policy_afinfo *afinfo = 2595 xfrm_policy_get_afinfo(xdst->u.dst.ops->family); 2596 int err; 2597 2598 if (!afinfo) 2599 return -EINVAL; 2600 2601 err = afinfo->fill_dst(xdst, dev, fl); 2602 2603 rcu_read_unlock(); 2604 2605 return err; 2606 } 2607 2608 2609 /* Allocate chain of dst_entry's, attach known xfrm's, calculate 2610 * all the metrics... Shortly, bundle a bundle. 2611 */ 2612 2613 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy, 2614 struct xfrm_state **xfrm, 2615 struct xfrm_dst **bundle, 2616 int nx, 2617 const struct flowi *fl, 2618 struct dst_entry *dst) 2619 { 2620 const struct xfrm_state_afinfo *afinfo; 2621 const struct xfrm_mode *inner_mode; 2622 struct net *net = xp_net(policy); 2623 unsigned long now = jiffies; 2624 struct net_device *dev; 2625 struct xfrm_dst *xdst_prev = NULL; 2626 struct xfrm_dst *xdst0 = NULL; 2627 int i = 0; 2628 int err; 2629 int header_len = 0; 2630 int nfheader_len = 0; 2631 int trailer_len = 0; 2632 int tos; 2633 int family = policy->selector.family; 2634 xfrm_address_t saddr, daddr; 2635 2636 xfrm_flowi_addr_get(fl, &saddr, &daddr, family); 2637 2638 tos = xfrm_get_tos(fl, family); 2639 2640 dst_hold(dst); 2641 2642 for (; i < nx; i++) { 2643 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family); 2644 struct dst_entry *dst1 = &xdst->u.dst; 2645 2646 err = PTR_ERR(xdst); 2647 if (IS_ERR(xdst)) { 2648 dst_release(dst); 2649 goto put_states; 2650 } 2651 2652 bundle[i] = xdst; 2653 if (!xdst_prev) 2654 xdst0 = xdst; 2655 else 2656 /* Ref count is taken during xfrm_alloc_dst() 2657 * No need to do dst_clone() on dst1 2658 */ 2659 xfrm_dst_set_child(xdst_prev, &xdst->u.dst); 2660 2661 if (xfrm[i]->sel.family == AF_UNSPEC) { 2662 inner_mode = xfrm_ip2inner_mode(xfrm[i], 2663 xfrm_af2proto(family)); 2664 if (!inner_mode) { 2665 err = -EAFNOSUPPORT; 2666 dst_release(dst); 2667 goto put_states; 2668 } 2669 } else 2670 inner_mode = &xfrm[i]->inner_mode; 2671 2672 xdst->route = dst; 2673 dst_copy_metrics(dst1, dst); 2674 2675 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) { 2676 __u32 mark = 0; 2677 int oif; 2678 2679 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m) 2680 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]); 2681 2682 if (xfrm[i]->xso.type != XFRM_DEV_OFFLOAD_PACKET) 2683 family = xfrm[i]->props.family; 2684 2685 oif = fl->flowi_oif ? : fl->flowi_l3mdev; 2686 dst = xfrm_dst_lookup(xfrm[i], tos, oif, 2687 &saddr, &daddr, family, mark); 2688 err = PTR_ERR(dst); 2689 if (IS_ERR(dst)) 2690 goto put_states; 2691 } else 2692 dst_hold(dst); 2693 2694 dst1->xfrm = xfrm[i]; 2695 xdst->xfrm_genid = xfrm[i]->genid; 2696 2697 dst1->obsolete = DST_OBSOLETE_FORCE_CHK; 2698 dst1->lastuse = now; 2699 2700 dst1->input = dst_discard; 2701 2702 rcu_read_lock(); 2703 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family); 2704 if (likely(afinfo)) 2705 dst1->output = afinfo->output; 2706 else 2707 dst1->output = dst_discard_out; 2708 rcu_read_unlock(); 2709 2710 xdst_prev = xdst; 2711 2712 header_len += xfrm[i]->props.header_len; 2713 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT) 2714 nfheader_len += xfrm[i]->props.header_len; 2715 trailer_len += xfrm[i]->props.trailer_len; 2716 } 2717 2718 xfrm_dst_set_child(xdst_prev, dst); 2719 xdst0->path = dst; 2720 2721 err = -ENODEV; 2722 dev = dst->dev; 2723 if (!dev) 2724 goto free_dst; 2725 2726 xfrm_init_path(xdst0, dst, nfheader_len); 2727 xfrm_init_pmtu(bundle, nx); 2728 2729 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst; 2730 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) { 2731 err = xfrm_fill_dst(xdst_prev, dev, fl); 2732 if (err) 2733 goto free_dst; 2734 2735 xdst_prev->u.dst.header_len = header_len; 2736 xdst_prev->u.dst.trailer_len = trailer_len; 2737 header_len -= xdst_prev->u.dst.xfrm->props.header_len; 2738 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len; 2739 } 2740 2741 return &xdst0->u.dst; 2742 2743 put_states: 2744 for (; i < nx; i++) 2745 xfrm_state_put(xfrm[i]); 2746 free_dst: 2747 if (xdst0) 2748 dst_release_immediate(&xdst0->u.dst); 2749 2750 return ERR_PTR(err); 2751 } 2752 2753 static int xfrm_expand_policies(const struct flowi *fl, u16 family, 2754 struct xfrm_policy **pols, 2755 int *num_pols, int *num_xfrms) 2756 { 2757 int i; 2758 2759 if (*num_pols == 0 || !pols[0]) { 2760 *num_pols = 0; 2761 *num_xfrms = 0; 2762 return 0; 2763 } 2764 if (IS_ERR(pols[0])) { 2765 *num_pols = 0; 2766 return PTR_ERR(pols[0]); 2767 } 2768 2769 *num_xfrms = pols[0]->xfrm_nr; 2770 2771 #ifdef CONFIG_XFRM_SUB_POLICY 2772 if (pols[0]->action == XFRM_POLICY_ALLOW && 2773 pols[0]->type != XFRM_POLICY_TYPE_MAIN) { 2774 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]), 2775 XFRM_POLICY_TYPE_MAIN, 2776 fl, family, 2777 XFRM_POLICY_OUT, 2778 pols[0]->if_id); 2779 if (pols[1]) { 2780 if (IS_ERR(pols[1])) { 2781 xfrm_pols_put(pols, *num_pols); 2782 *num_pols = 0; 2783 return PTR_ERR(pols[1]); 2784 } 2785 (*num_pols)++; 2786 (*num_xfrms) += pols[1]->xfrm_nr; 2787 } 2788 } 2789 #endif 2790 for (i = 0; i < *num_pols; i++) { 2791 if (pols[i]->action != XFRM_POLICY_ALLOW) { 2792 *num_xfrms = -1; 2793 break; 2794 } 2795 } 2796 2797 return 0; 2798 2799 } 2800 2801 static struct xfrm_dst * 2802 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols, 2803 const struct flowi *fl, u16 family, 2804 struct dst_entry *dst_orig) 2805 { 2806 struct net *net = xp_net(pols[0]); 2807 struct xfrm_state *xfrm[XFRM_MAX_DEPTH]; 2808 struct xfrm_dst *bundle[XFRM_MAX_DEPTH]; 2809 struct xfrm_dst *xdst; 2810 struct dst_entry *dst; 2811 int err; 2812 2813 /* Try to instantiate a bundle */ 2814 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family); 2815 if (err <= 0) { 2816 if (err == 0) 2817 return NULL; 2818 2819 if (err != -EAGAIN) 2820 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR); 2821 return ERR_PTR(err); 2822 } 2823 2824 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig); 2825 if (IS_ERR(dst)) { 2826 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR); 2827 return ERR_CAST(dst); 2828 } 2829 2830 xdst = (struct xfrm_dst *)dst; 2831 xdst->num_xfrms = err; 2832 xdst->num_pols = num_pols; 2833 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols); 2834 xdst->policy_genid = atomic_read(&pols[0]->genid); 2835 2836 return xdst; 2837 } 2838 2839 static void xfrm_policy_queue_process(struct timer_list *t) 2840 { 2841 struct sk_buff *skb; 2842 struct sock *sk; 2843 struct dst_entry *dst; 2844 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer); 2845 struct net *net = xp_net(pol); 2846 struct xfrm_policy_queue *pq = &pol->polq; 2847 struct flowi fl; 2848 struct sk_buff_head list; 2849 __u32 skb_mark; 2850 2851 spin_lock(&pq->hold_queue.lock); 2852 skb = skb_peek(&pq->hold_queue); 2853 if (!skb) { 2854 spin_unlock(&pq->hold_queue.lock); 2855 goto out; 2856 } 2857 dst = skb_dst(skb); 2858 sk = skb->sk; 2859 2860 /* Fixup the mark to support VTI. */ 2861 skb_mark = skb->mark; 2862 skb->mark = pol->mark.v; 2863 xfrm_decode_session(skb, &fl, dst->ops->family); 2864 skb->mark = skb_mark; 2865 spin_unlock(&pq->hold_queue.lock); 2866 2867 dst_hold(xfrm_dst_path(dst)); 2868 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE); 2869 if (IS_ERR(dst)) 2870 goto purge_queue; 2871 2872 if (dst->flags & DST_XFRM_QUEUE) { 2873 dst_release(dst); 2874 2875 if (pq->timeout >= XFRM_QUEUE_TMO_MAX) 2876 goto purge_queue; 2877 2878 pq->timeout = pq->timeout << 1; 2879 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout)) 2880 xfrm_pol_hold(pol); 2881 goto out; 2882 } 2883 2884 dst_release(dst); 2885 2886 __skb_queue_head_init(&list); 2887 2888 spin_lock(&pq->hold_queue.lock); 2889 pq->timeout = 0; 2890 skb_queue_splice_init(&pq->hold_queue, &list); 2891 spin_unlock(&pq->hold_queue.lock); 2892 2893 while (!skb_queue_empty(&list)) { 2894 skb = __skb_dequeue(&list); 2895 2896 /* Fixup the mark to support VTI. */ 2897 skb_mark = skb->mark; 2898 skb->mark = pol->mark.v; 2899 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family); 2900 skb->mark = skb_mark; 2901 2902 dst_hold(xfrm_dst_path(skb_dst(skb))); 2903 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0); 2904 if (IS_ERR(dst)) { 2905 kfree_skb(skb); 2906 continue; 2907 } 2908 2909 nf_reset_ct(skb); 2910 skb_dst_drop(skb); 2911 skb_dst_set(skb, dst); 2912 2913 dst_output(net, skb->sk, skb); 2914 } 2915 2916 out: 2917 xfrm_pol_put(pol); 2918 return; 2919 2920 purge_queue: 2921 pq->timeout = 0; 2922 skb_queue_purge(&pq->hold_queue); 2923 xfrm_pol_put(pol); 2924 } 2925 2926 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb) 2927 { 2928 unsigned long sched_next; 2929 struct dst_entry *dst = skb_dst(skb); 2930 struct xfrm_dst *xdst = (struct xfrm_dst *) dst; 2931 struct xfrm_policy *pol = xdst->pols[0]; 2932 struct xfrm_policy_queue *pq = &pol->polq; 2933 2934 if (unlikely(skb_fclone_busy(sk, skb))) { 2935 kfree_skb(skb); 2936 return 0; 2937 } 2938 2939 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) { 2940 kfree_skb(skb); 2941 return -EAGAIN; 2942 } 2943 2944 skb_dst_force(skb); 2945 2946 spin_lock_bh(&pq->hold_queue.lock); 2947 2948 if (!pq->timeout) 2949 pq->timeout = XFRM_QUEUE_TMO_MIN; 2950 2951 sched_next = jiffies + pq->timeout; 2952 2953 if (del_timer(&pq->hold_timer)) { 2954 if (time_before(pq->hold_timer.expires, sched_next)) 2955 sched_next = pq->hold_timer.expires; 2956 xfrm_pol_put(pol); 2957 } 2958 2959 __skb_queue_tail(&pq->hold_queue, skb); 2960 if (!mod_timer(&pq->hold_timer, sched_next)) 2961 xfrm_pol_hold(pol); 2962 2963 spin_unlock_bh(&pq->hold_queue.lock); 2964 2965 return 0; 2966 } 2967 2968 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net, 2969 struct xfrm_flo *xflo, 2970 const struct flowi *fl, 2971 int num_xfrms, 2972 u16 family) 2973 { 2974 int err; 2975 struct net_device *dev; 2976 struct dst_entry *dst; 2977 struct dst_entry *dst1; 2978 struct xfrm_dst *xdst; 2979 2980 xdst = xfrm_alloc_dst(net, family); 2981 if (IS_ERR(xdst)) 2982 return xdst; 2983 2984 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) || 2985 net->xfrm.sysctl_larval_drop || 2986 num_xfrms <= 0) 2987 return xdst; 2988 2989 dst = xflo->dst_orig; 2990 dst1 = &xdst->u.dst; 2991 dst_hold(dst); 2992 xdst->route = dst; 2993 2994 dst_copy_metrics(dst1, dst); 2995 2996 dst1->obsolete = DST_OBSOLETE_FORCE_CHK; 2997 dst1->flags |= DST_XFRM_QUEUE; 2998 dst1->lastuse = jiffies; 2999 3000 dst1->input = dst_discard; 3001 dst1->output = xdst_queue_output; 3002 3003 dst_hold(dst); 3004 xfrm_dst_set_child(xdst, dst); 3005 xdst->path = dst; 3006 3007 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0); 3008 3009 err = -ENODEV; 3010 dev = dst->dev; 3011 if (!dev) 3012 goto free_dst; 3013 3014 err = xfrm_fill_dst(xdst, dev, fl); 3015 if (err) 3016 goto free_dst; 3017 3018 out: 3019 return xdst; 3020 3021 free_dst: 3022 dst_release(dst1); 3023 xdst = ERR_PTR(err); 3024 goto out; 3025 } 3026 3027 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net, 3028 const struct flowi *fl, 3029 u16 family, u8 dir, 3030 struct xfrm_flo *xflo, u32 if_id) 3031 { 3032 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 3033 int num_pols = 0, num_xfrms = 0, err; 3034 struct xfrm_dst *xdst; 3035 3036 /* Resolve policies to use if we couldn't get them from 3037 * previous cache entry */ 3038 num_pols = 1; 3039 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id); 3040 err = xfrm_expand_policies(fl, family, pols, 3041 &num_pols, &num_xfrms); 3042 if (err < 0) 3043 goto inc_error; 3044 if (num_pols == 0) 3045 return NULL; 3046 if (num_xfrms <= 0) 3047 goto make_dummy_bundle; 3048 3049 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, 3050 xflo->dst_orig); 3051 if (IS_ERR(xdst)) { 3052 err = PTR_ERR(xdst); 3053 if (err == -EREMOTE) { 3054 xfrm_pols_put(pols, num_pols); 3055 return NULL; 3056 } 3057 3058 if (err != -EAGAIN) 3059 goto error; 3060 goto make_dummy_bundle; 3061 } else if (xdst == NULL) { 3062 num_xfrms = 0; 3063 goto make_dummy_bundle; 3064 } 3065 3066 return xdst; 3067 3068 make_dummy_bundle: 3069 /* We found policies, but there's no bundles to instantiate: 3070 * either because the policy blocks, has no transformations or 3071 * we could not build template (no xfrm_states).*/ 3072 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family); 3073 if (IS_ERR(xdst)) { 3074 xfrm_pols_put(pols, num_pols); 3075 return ERR_CAST(xdst); 3076 } 3077 xdst->num_pols = num_pols; 3078 xdst->num_xfrms = num_xfrms; 3079 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols); 3080 3081 return xdst; 3082 3083 inc_error: 3084 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR); 3085 error: 3086 xfrm_pols_put(pols, num_pols); 3087 return ERR_PTR(err); 3088 } 3089 3090 static struct dst_entry *make_blackhole(struct net *net, u16 family, 3091 struct dst_entry *dst_orig) 3092 { 3093 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 3094 struct dst_entry *ret; 3095 3096 if (!afinfo) { 3097 dst_release(dst_orig); 3098 return ERR_PTR(-EINVAL); 3099 } else { 3100 ret = afinfo->blackhole_route(net, dst_orig); 3101 } 3102 rcu_read_unlock(); 3103 3104 return ret; 3105 } 3106 3107 /* Finds/creates a bundle for given flow and if_id 3108 * 3109 * At the moment we eat a raw IP route. Mostly to speed up lookups 3110 * on interfaces with disabled IPsec. 3111 * 3112 * xfrm_lookup uses an if_id of 0 by default, and is provided for 3113 * compatibility 3114 */ 3115 struct dst_entry *xfrm_lookup_with_ifid(struct net *net, 3116 struct dst_entry *dst_orig, 3117 const struct flowi *fl, 3118 const struct sock *sk, 3119 int flags, u32 if_id) 3120 { 3121 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 3122 struct xfrm_dst *xdst; 3123 struct dst_entry *dst, *route; 3124 u16 family = dst_orig->ops->family; 3125 u8 dir = XFRM_POLICY_OUT; 3126 int i, err, num_pols, num_xfrms = 0, drop_pols = 0; 3127 3128 dst = NULL; 3129 xdst = NULL; 3130 route = NULL; 3131 3132 sk = sk_const_to_full_sk(sk); 3133 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) { 3134 num_pols = 1; 3135 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family, 3136 if_id); 3137 err = xfrm_expand_policies(fl, family, pols, 3138 &num_pols, &num_xfrms); 3139 if (err < 0) 3140 goto dropdst; 3141 3142 if (num_pols) { 3143 if (num_xfrms <= 0) { 3144 drop_pols = num_pols; 3145 goto no_transform; 3146 } 3147 3148 xdst = xfrm_resolve_and_create_bundle( 3149 pols, num_pols, fl, 3150 family, dst_orig); 3151 3152 if (IS_ERR(xdst)) { 3153 xfrm_pols_put(pols, num_pols); 3154 err = PTR_ERR(xdst); 3155 if (err == -EREMOTE) 3156 goto nopol; 3157 3158 goto dropdst; 3159 } else if (xdst == NULL) { 3160 num_xfrms = 0; 3161 drop_pols = num_pols; 3162 goto no_transform; 3163 } 3164 3165 route = xdst->route; 3166 } 3167 } 3168 3169 if (xdst == NULL) { 3170 struct xfrm_flo xflo; 3171 3172 xflo.dst_orig = dst_orig; 3173 xflo.flags = flags; 3174 3175 /* To accelerate a bit... */ 3176 if (!if_id && ((dst_orig->flags & DST_NOXFRM) || 3177 !net->xfrm.policy_count[XFRM_POLICY_OUT])) 3178 goto nopol; 3179 3180 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id); 3181 if (xdst == NULL) 3182 goto nopol; 3183 if (IS_ERR(xdst)) { 3184 err = PTR_ERR(xdst); 3185 goto dropdst; 3186 } 3187 3188 num_pols = xdst->num_pols; 3189 num_xfrms = xdst->num_xfrms; 3190 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols); 3191 route = xdst->route; 3192 } 3193 3194 dst = &xdst->u.dst; 3195 if (route == NULL && num_xfrms > 0) { 3196 /* The only case when xfrm_bundle_lookup() returns a 3197 * bundle with null route, is when the template could 3198 * not be resolved. It means policies are there, but 3199 * bundle could not be created, since we don't yet 3200 * have the xfrm_state's. We need to wait for KM to 3201 * negotiate new SA's or bail out with error.*/ 3202 if (net->xfrm.sysctl_larval_drop) { 3203 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES); 3204 err = -EREMOTE; 3205 goto error; 3206 } 3207 3208 err = -EAGAIN; 3209 3210 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES); 3211 goto error; 3212 } 3213 3214 no_transform: 3215 if (num_pols == 0) 3216 goto nopol; 3217 3218 if ((flags & XFRM_LOOKUP_ICMP) && 3219 !(pols[0]->flags & XFRM_POLICY_ICMP)) { 3220 err = -ENOENT; 3221 goto error; 3222 } 3223 3224 for (i = 0; i < num_pols; i++) 3225 WRITE_ONCE(pols[i]->curlft.use_time, ktime_get_real_seconds()); 3226 3227 if (num_xfrms < 0) { 3228 /* Prohibit the flow */ 3229 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK); 3230 err = -EPERM; 3231 goto error; 3232 } else if (num_xfrms > 0) { 3233 /* Flow transformed */ 3234 dst_release(dst_orig); 3235 } else { 3236 /* Flow passes untransformed */ 3237 dst_release(dst); 3238 dst = dst_orig; 3239 } 3240 ok: 3241 xfrm_pols_put(pols, drop_pols); 3242 if (dst && dst->xfrm && 3243 dst->xfrm->props.mode == XFRM_MODE_TUNNEL) 3244 dst->flags |= DST_XFRM_TUNNEL; 3245 return dst; 3246 3247 nopol: 3248 if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) && 3249 net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) { 3250 err = -EPERM; 3251 goto error; 3252 } 3253 if (!(flags & XFRM_LOOKUP_ICMP)) { 3254 dst = dst_orig; 3255 goto ok; 3256 } 3257 err = -ENOENT; 3258 error: 3259 dst_release(dst); 3260 dropdst: 3261 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF)) 3262 dst_release(dst_orig); 3263 xfrm_pols_put(pols, drop_pols); 3264 return ERR_PTR(err); 3265 } 3266 EXPORT_SYMBOL(xfrm_lookup_with_ifid); 3267 3268 /* Main function: finds/creates a bundle for given flow. 3269 * 3270 * At the moment we eat a raw IP route. Mostly to speed up lookups 3271 * on interfaces with disabled IPsec. 3272 */ 3273 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig, 3274 const struct flowi *fl, const struct sock *sk, 3275 int flags) 3276 { 3277 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0); 3278 } 3279 EXPORT_SYMBOL(xfrm_lookup); 3280 3281 /* Callers of xfrm_lookup_route() must ensure a call to dst_output(). 3282 * Otherwise we may send out blackholed packets. 3283 */ 3284 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig, 3285 const struct flowi *fl, 3286 const struct sock *sk, int flags) 3287 { 3288 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk, 3289 flags | XFRM_LOOKUP_QUEUE | 3290 XFRM_LOOKUP_KEEP_DST_REF); 3291 3292 if (PTR_ERR(dst) == -EREMOTE) 3293 return make_blackhole(net, dst_orig->ops->family, dst_orig); 3294 3295 if (IS_ERR(dst)) 3296 dst_release(dst_orig); 3297 3298 return dst; 3299 } 3300 EXPORT_SYMBOL(xfrm_lookup_route); 3301 3302 static inline int 3303 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl) 3304 { 3305 struct sec_path *sp = skb_sec_path(skb); 3306 struct xfrm_state *x; 3307 3308 if (!sp || idx < 0 || idx >= sp->len) 3309 return 0; 3310 x = sp->xvec[idx]; 3311 if (!x->type->reject) 3312 return 0; 3313 return x->type->reject(x, skb, fl); 3314 } 3315 3316 /* When skb is transformed back to its "native" form, we have to 3317 * check policy restrictions. At the moment we make this in maximally 3318 * stupid way. Shame on me. :-) Of course, connected sockets must 3319 * have policy cached at them. 3320 */ 3321 3322 static inline int 3323 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x, 3324 unsigned short family, u32 if_id) 3325 { 3326 if (xfrm_state_kern(x)) 3327 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family); 3328 return x->id.proto == tmpl->id.proto && 3329 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) && 3330 (x->props.reqid == tmpl->reqid || !tmpl->reqid) && 3331 x->props.mode == tmpl->mode && 3332 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) || 3333 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) && 3334 !(x->props.mode != XFRM_MODE_TRANSPORT && 3335 xfrm_state_addr_cmp(tmpl, x, family)) && 3336 (if_id == 0 || if_id == x->if_id); 3337 } 3338 3339 /* 3340 * 0 or more than 0 is returned when validation is succeeded (either bypass 3341 * because of optional transport mode, or next index of the matched secpath 3342 * state with the template. 3343 * -1 is returned when no matching template is found. 3344 * Otherwise "-2 - errored_index" is returned. 3345 */ 3346 static inline int 3347 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start, 3348 unsigned short family, u32 if_id) 3349 { 3350 int idx = start; 3351 3352 if (tmpl->optional) { 3353 if (tmpl->mode == XFRM_MODE_TRANSPORT) 3354 return start; 3355 } else 3356 start = -1; 3357 for (; idx < sp->len; idx++) { 3358 if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id)) 3359 return ++idx; 3360 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) { 3361 if (idx < sp->verified_cnt) { 3362 /* Secpath entry previously verified, consider optional and 3363 * continue searching 3364 */ 3365 continue; 3366 } 3367 3368 if (start == -1) 3369 start = -2-idx; 3370 break; 3371 } 3372 } 3373 return start; 3374 } 3375 3376 static void 3377 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse) 3378 { 3379 const struct iphdr *iph = ip_hdr(skb); 3380 int ihl = iph->ihl; 3381 u8 *xprth = skb_network_header(skb) + ihl * 4; 3382 struct flowi4 *fl4 = &fl->u.ip4; 3383 int oif = 0; 3384 3385 if (skb_dst(skb) && skb_dst(skb)->dev) 3386 oif = skb_dst(skb)->dev->ifindex; 3387 3388 memset(fl4, 0, sizeof(struct flowi4)); 3389 fl4->flowi4_mark = skb->mark; 3390 fl4->flowi4_oif = reverse ? skb->skb_iif : oif; 3391 3392 fl4->flowi4_proto = iph->protocol; 3393 fl4->daddr = reverse ? iph->saddr : iph->daddr; 3394 fl4->saddr = reverse ? iph->daddr : iph->saddr; 3395 fl4->flowi4_tos = iph->tos & ~INET_ECN_MASK; 3396 3397 if (!ip_is_fragment(iph)) { 3398 switch (iph->protocol) { 3399 case IPPROTO_UDP: 3400 case IPPROTO_UDPLITE: 3401 case IPPROTO_TCP: 3402 case IPPROTO_SCTP: 3403 case IPPROTO_DCCP: 3404 if (xprth + 4 < skb->data || 3405 pskb_may_pull(skb, xprth + 4 - skb->data)) { 3406 __be16 *ports; 3407 3408 xprth = skb_network_header(skb) + ihl * 4; 3409 ports = (__be16 *)xprth; 3410 3411 fl4->fl4_sport = ports[!!reverse]; 3412 fl4->fl4_dport = ports[!reverse]; 3413 } 3414 break; 3415 case IPPROTO_ICMP: 3416 if (xprth + 2 < skb->data || 3417 pskb_may_pull(skb, xprth + 2 - skb->data)) { 3418 u8 *icmp; 3419 3420 xprth = skb_network_header(skb) + ihl * 4; 3421 icmp = xprth; 3422 3423 fl4->fl4_icmp_type = icmp[0]; 3424 fl4->fl4_icmp_code = icmp[1]; 3425 } 3426 break; 3427 case IPPROTO_GRE: 3428 if (xprth + 12 < skb->data || 3429 pskb_may_pull(skb, xprth + 12 - skb->data)) { 3430 __be16 *greflags; 3431 __be32 *gre_hdr; 3432 3433 xprth = skb_network_header(skb) + ihl * 4; 3434 greflags = (__be16 *)xprth; 3435 gre_hdr = (__be32 *)xprth; 3436 3437 if (greflags[0] & GRE_KEY) { 3438 if (greflags[0] & GRE_CSUM) 3439 gre_hdr++; 3440 fl4->fl4_gre_key = gre_hdr[1]; 3441 } 3442 } 3443 break; 3444 default: 3445 break; 3446 } 3447 } 3448 } 3449 3450 #if IS_ENABLED(CONFIG_IPV6) 3451 static void 3452 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse) 3453 { 3454 struct flowi6 *fl6 = &fl->u.ip6; 3455 int onlyproto = 0; 3456 const struct ipv6hdr *hdr = ipv6_hdr(skb); 3457 u32 offset = sizeof(*hdr); 3458 struct ipv6_opt_hdr *exthdr; 3459 const unsigned char *nh = skb_network_header(skb); 3460 u16 nhoff = IP6CB(skb)->nhoff; 3461 int oif = 0; 3462 u8 nexthdr; 3463 3464 if (!nhoff) 3465 nhoff = offsetof(struct ipv6hdr, nexthdr); 3466 3467 nexthdr = nh[nhoff]; 3468 3469 if (skb_dst(skb) && skb_dst(skb)->dev) 3470 oif = skb_dst(skb)->dev->ifindex; 3471 3472 memset(fl6, 0, sizeof(struct flowi6)); 3473 fl6->flowi6_mark = skb->mark; 3474 fl6->flowi6_oif = reverse ? skb->skb_iif : oif; 3475 3476 fl6->daddr = reverse ? hdr->saddr : hdr->daddr; 3477 fl6->saddr = reverse ? hdr->daddr : hdr->saddr; 3478 3479 while (nh + offset + sizeof(*exthdr) < skb->data || 3480 pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) { 3481 nh = skb_network_header(skb); 3482 exthdr = (struct ipv6_opt_hdr *)(nh + offset); 3483 3484 switch (nexthdr) { 3485 case NEXTHDR_FRAGMENT: 3486 onlyproto = 1; 3487 fallthrough; 3488 case NEXTHDR_ROUTING: 3489 case NEXTHDR_HOP: 3490 case NEXTHDR_DEST: 3491 offset += ipv6_optlen(exthdr); 3492 nexthdr = exthdr->nexthdr; 3493 break; 3494 case IPPROTO_UDP: 3495 case IPPROTO_UDPLITE: 3496 case IPPROTO_TCP: 3497 case IPPROTO_SCTP: 3498 case IPPROTO_DCCP: 3499 if (!onlyproto && (nh + offset + 4 < skb->data || 3500 pskb_may_pull(skb, nh + offset + 4 - skb->data))) { 3501 __be16 *ports; 3502 3503 nh = skb_network_header(skb); 3504 ports = (__be16 *)(nh + offset); 3505 fl6->fl6_sport = ports[!!reverse]; 3506 fl6->fl6_dport = ports[!reverse]; 3507 } 3508 fl6->flowi6_proto = nexthdr; 3509 return; 3510 case IPPROTO_ICMPV6: 3511 if (!onlyproto && (nh + offset + 2 < skb->data || 3512 pskb_may_pull(skb, nh + offset + 2 - skb->data))) { 3513 u8 *icmp; 3514 3515 nh = skb_network_header(skb); 3516 icmp = (u8 *)(nh + offset); 3517 fl6->fl6_icmp_type = icmp[0]; 3518 fl6->fl6_icmp_code = icmp[1]; 3519 } 3520 fl6->flowi6_proto = nexthdr; 3521 return; 3522 case IPPROTO_GRE: 3523 if (!onlyproto && 3524 (nh + offset + 12 < skb->data || 3525 pskb_may_pull(skb, nh + offset + 12 - skb->data))) { 3526 struct gre_base_hdr *gre_hdr; 3527 __be32 *gre_key; 3528 3529 nh = skb_network_header(skb); 3530 gre_hdr = (struct gre_base_hdr *)(nh + offset); 3531 gre_key = (__be32 *)(gre_hdr + 1); 3532 3533 if (gre_hdr->flags & GRE_KEY) { 3534 if (gre_hdr->flags & GRE_CSUM) 3535 gre_key++; 3536 fl6->fl6_gre_key = *gre_key; 3537 } 3538 } 3539 fl6->flowi6_proto = nexthdr; 3540 return; 3541 3542 #if IS_ENABLED(CONFIG_IPV6_MIP6) 3543 case IPPROTO_MH: 3544 offset += ipv6_optlen(exthdr); 3545 if (!onlyproto && (nh + offset + 3 < skb->data || 3546 pskb_may_pull(skb, nh + offset + 3 - skb->data))) { 3547 struct ip6_mh *mh; 3548 3549 nh = skb_network_header(skb); 3550 mh = (struct ip6_mh *)(nh + offset); 3551 fl6->fl6_mh_type = mh->ip6mh_type; 3552 } 3553 fl6->flowi6_proto = nexthdr; 3554 return; 3555 #endif 3556 default: 3557 fl6->flowi6_proto = nexthdr; 3558 return; 3559 } 3560 } 3561 } 3562 #endif 3563 3564 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, 3565 unsigned int family, int reverse) 3566 { 3567 switch (family) { 3568 case AF_INET: 3569 decode_session4(skb, fl, reverse); 3570 break; 3571 #if IS_ENABLED(CONFIG_IPV6) 3572 case AF_INET6: 3573 decode_session6(skb, fl, reverse); 3574 break; 3575 #endif 3576 default: 3577 return -EAFNOSUPPORT; 3578 } 3579 3580 return security_xfrm_decode_session(skb, &fl->flowi_secid); 3581 } 3582 EXPORT_SYMBOL(__xfrm_decode_session); 3583 3584 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp) 3585 { 3586 for (; k < sp->len; k++) { 3587 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) { 3588 *idxp = k; 3589 return 1; 3590 } 3591 } 3592 3593 return 0; 3594 } 3595 3596 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 3597 unsigned short family) 3598 { 3599 struct net *net = dev_net(skb->dev); 3600 struct xfrm_policy *pol; 3601 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 3602 int npols = 0; 3603 int xfrm_nr; 3604 int pi; 3605 int reverse; 3606 struct flowi fl; 3607 int xerr_idx = -1; 3608 const struct xfrm_if_cb *ifcb; 3609 struct sec_path *sp; 3610 u32 if_id = 0; 3611 3612 rcu_read_lock(); 3613 ifcb = xfrm_if_get_cb(); 3614 3615 if (ifcb) { 3616 struct xfrm_if_decode_session_result r; 3617 3618 if (ifcb->decode_session(skb, family, &r)) { 3619 if_id = r.if_id; 3620 net = r.net; 3621 } 3622 } 3623 rcu_read_unlock(); 3624 3625 reverse = dir & ~XFRM_POLICY_MASK; 3626 dir &= XFRM_POLICY_MASK; 3627 3628 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) { 3629 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR); 3630 return 0; 3631 } 3632 3633 nf_nat_decode_session(skb, &fl, family); 3634 3635 /* First, check used SA against their selectors. */ 3636 sp = skb_sec_path(skb); 3637 if (sp) { 3638 int i; 3639 3640 for (i = sp->len - 1; i >= 0; i--) { 3641 struct xfrm_state *x = sp->xvec[i]; 3642 if (!xfrm_selector_match(&x->sel, &fl, family)) { 3643 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH); 3644 return 0; 3645 } 3646 } 3647 } 3648 3649 pol = NULL; 3650 sk = sk_to_full_sk(sk); 3651 if (sk && sk->sk_policy[dir]) { 3652 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id); 3653 if (IS_ERR(pol)) { 3654 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3655 return 0; 3656 } 3657 } 3658 3659 if (!pol) 3660 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id); 3661 3662 if (IS_ERR(pol)) { 3663 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3664 return 0; 3665 } 3666 3667 if (!pol) { 3668 if (net->xfrm.policy_default[dir] == XFRM_USERPOLICY_BLOCK) { 3669 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS); 3670 return 0; 3671 } 3672 3673 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) { 3674 xfrm_secpath_reject(xerr_idx, skb, &fl); 3675 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS); 3676 return 0; 3677 } 3678 return 1; 3679 } 3680 3681 /* This lockless write can happen from different cpus. */ 3682 WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds()); 3683 3684 pols[0] = pol; 3685 npols++; 3686 #ifdef CONFIG_XFRM_SUB_POLICY 3687 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) { 3688 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, 3689 &fl, family, 3690 XFRM_POLICY_IN, if_id); 3691 if (pols[1]) { 3692 if (IS_ERR(pols[1])) { 3693 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3694 xfrm_pol_put(pols[0]); 3695 return 0; 3696 } 3697 /* This write can happen from different cpus. */ 3698 WRITE_ONCE(pols[1]->curlft.use_time, 3699 ktime_get_real_seconds()); 3700 npols++; 3701 } 3702 } 3703 #endif 3704 3705 if (pol->action == XFRM_POLICY_ALLOW) { 3706 static struct sec_path dummy; 3707 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH]; 3708 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH]; 3709 struct xfrm_tmpl **tpp = tp; 3710 int ti = 0; 3711 int i, k; 3712 3713 sp = skb_sec_path(skb); 3714 if (!sp) 3715 sp = &dummy; 3716 3717 for (pi = 0; pi < npols; pi++) { 3718 if (pols[pi] != pol && 3719 pols[pi]->action != XFRM_POLICY_ALLOW) { 3720 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK); 3721 goto reject; 3722 } 3723 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) { 3724 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR); 3725 goto reject_error; 3726 } 3727 for (i = 0; i < pols[pi]->xfrm_nr; i++) 3728 tpp[ti++] = &pols[pi]->xfrm_vec[i]; 3729 } 3730 xfrm_nr = ti; 3731 3732 if (npols > 1) { 3733 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family); 3734 tpp = stp; 3735 } 3736 3737 /* For each tunnel xfrm, find the first matching tmpl. 3738 * For each tmpl before that, find corresponding xfrm. 3739 * Order is _important_. Later we will implement 3740 * some barriers, but at the moment barriers 3741 * are implied between each two transformations. 3742 * Upon success, marks secpath entries as having been 3743 * verified to allow them to be skipped in future policy 3744 * checks (e.g. nested tunnels). 3745 */ 3746 for (i = xfrm_nr-1, k = 0; i >= 0; i--) { 3747 k = xfrm_policy_ok(tpp[i], sp, k, family, if_id); 3748 if (k < 0) { 3749 if (k < -1) 3750 /* "-2 - errored_index" returned */ 3751 xerr_idx = -(2+k); 3752 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH); 3753 goto reject; 3754 } 3755 } 3756 3757 if (secpath_has_nontransport(sp, k, &xerr_idx)) { 3758 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH); 3759 goto reject; 3760 } 3761 3762 xfrm_pols_put(pols, npols); 3763 sp->verified_cnt = k; 3764 3765 return 1; 3766 } 3767 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK); 3768 3769 reject: 3770 xfrm_secpath_reject(xerr_idx, skb, &fl); 3771 reject_error: 3772 xfrm_pols_put(pols, npols); 3773 return 0; 3774 } 3775 EXPORT_SYMBOL(__xfrm_policy_check); 3776 3777 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family) 3778 { 3779 struct net *net = dev_net(skb->dev); 3780 struct flowi fl; 3781 struct dst_entry *dst; 3782 int res = 1; 3783 3784 if (xfrm_decode_session(skb, &fl, family) < 0) { 3785 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR); 3786 return 0; 3787 } 3788 3789 skb_dst_force(skb); 3790 if (!skb_dst(skb)) { 3791 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR); 3792 return 0; 3793 } 3794 3795 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE); 3796 if (IS_ERR(dst)) { 3797 res = 0; 3798 dst = NULL; 3799 } 3800 skb_dst_set(skb, dst); 3801 return res; 3802 } 3803 EXPORT_SYMBOL(__xfrm_route_forward); 3804 3805 /* Optimize later using cookies and generation ids. */ 3806 3807 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie) 3808 { 3809 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete 3810 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to 3811 * get validated by dst_ops->check on every use. We do this 3812 * because when a normal route referenced by an XFRM dst is 3813 * obsoleted we do not go looking around for all parent 3814 * referencing XFRM dsts so that we can invalidate them. It 3815 * is just too much work. Instead we make the checks here on 3816 * every use. For example: 3817 * 3818 * XFRM dst A --> IPv4 dst X 3819 * 3820 * X is the "xdst->route" of A (X is also the "dst->path" of A 3821 * in this example). If X is marked obsolete, "A" will not 3822 * notice. That's what we are validating here via the 3823 * stale_bundle() check. 3824 * 3825 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will 3826 * be marked on it. 3827 * This will force stale_bundle() to fail on any xdst bundle with 3828 * this dst linked in it. 3829 */ 3830 if (dst->obsolete < 0 && !stale_bundle(dst)) 3831 return dst; 3832 3833 return NULL; 3834 } 3835 3836 static int stale_bundle(struct dst_entry *dst) 3837 { 3838 return !xfrm_bundle_ok((struct xfrm_dst *)dst); 3839 } 3840 3841 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev) 3842 { 3843 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) { 3844 dst->dev = blackhole_netdev; 3845 dev_hold(dst->dev); 3846 dev_put(dev); 3847 } 3848 } 3849 EXPORT_SYMBOL(xfrm_dst_ifdown); 3850 3851 static void xfrm_link_failure(struct sk_buff *skb) 3852 { 3853 /* Impossible. Such dst must be popped before reaches point of failure. */ 3854 } 3855 3856 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst) 3857 { 3858 if (dst) { 3859 if (dst->obsolete) { 3860 dst_release(dst); 3861 dst = NULL; 3862 } 3863 } 3864 return dst; 3865 } 3866 3867 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr) 3868 { 3869 while (nr--) { 3870 struct xfrm_dst *xdst = bundle[nr]; 3871 u32 pmtu, route_mtu_cached; 3872 struct dst_entry *dst; 3873 3874 dst = &xdst->u.dst; 3875 pmtu = dst_mtu(xfrm_dst_child(dst)); 3876 xdst->child_mtu_cached = pmtu; 3877 3878 pmtu = xfrm_state_mtu(dst->xfrm, pmtu); 3879 3880 route_mtu_cached = dst_mtu(xdst->route); 3881 xdst->route_mtu_cached = route_mtu_cached; 3882 3883 if (pmtu > route_mtu_cached) 3884 pmtu = route_mtu_cached; 3885 3886 dst_metric_set(dst, RTAX_MTU, pmtu); 3887 } 3888 } 3889 3890 /* Check that the bundle accepts the flow and its components are 3891 * still valid. 3892 */ 3893 3894 static int xfrm_bundle_ok(struct xfrm_dst *first) 3895 { 3896 struct xfrm_dst *bundle[XFRM_MAX_DEPTH]; 3897 struct dst_entry *dst = &first->u.dst; 3898 struct xfrm_dst *xdst; 3899 int start_from, nr; 3900 u32 mtu; 3901 3902 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) || 3903 (dst->dev && !netif_running(dst->dev))) 3904 return 0; 3905 3906 if (dst->flags & DST_XFRM_QUEUE) 3907 return 1; 3908 3909 start_from = nr = 0; 3910 do { 3911 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 3912 3913 if (dst->xfrm->km.state != XFRM_STATE_VALID) 3914 return 0; 3915 if (xdst->xfrm_genid != dst->xfrm->genid) 3916 return 0; 3917 if (xdst->num_pols > 0 && 3918 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid)) 3919 return 0; 3920 3921 bundle[nr++] = xdst; 3922 3923 mtu = dst_mtu(xfrm_dst_child(dst)); 3924 if (xdst->child_mtu_cached != mtu) { 3925 start_from = nr; 3926 xdst->child_mtu_cached = mtu; 3927 } 3928 3929 if (!dst_check(xdst->route, xdst->route_cookie)) 3930 return 0; 3931 mtu = dst_mtu(xdst->route); 3932 if (xdst->route_mtu_cached != mtu) { 3933 start_from = nr; 3934 xdst->route_mtu_cached = mtu; 3935 } 3936 3937 dst = xfrm_dst_child(dst); 3938 } while (dst->xfrm); 3939 3940 if (likely(!start_from)) 3941 return 1; 3942 3943 xdst = bundle[start_from - 1]; 3944 mtu = xdst->child_mtu_cached; 3945 while (start_from--) { 3946 dst = &xdst->u.dst; 3947 3948 mtu = xfrm_state_mtu(dst->xfrm, mtu); 3949 if (mtu > xdst->route_mtu_cached) 3950 mtu = xdst->route_mtu_cached; 3951 dst_metric_set(dst, RTAX_MTU, mtu); 3952 if (!start_from) 3953 break; 3954 3955 xdst = bundle[start_from - 1]; 3956 xdst->child_mtu_cached = mtu; 3957 } 3958 3959 return 1; 3960 } 3961 3962 static unsigned int xfrm_default_advmss(const struct dst_entry *dst) 3963 { 3964 return dst_metric_advmss(xfrm_dst_path(dst)); 3965 } 3966 3967 static unsigned int xfrm_mtu(const struct dst_entry *dst) 3968 { 3969 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 3970 3971 return mtu ? : dst_mtu(xfrm_dst_path(dst)); 3972 } 3973 3974 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst, 3975 const void *daddr) 3976 { 3977 while (dst->xfrm) { 3978 const struct xfrm_state *xfrm = dst->xfrm; 3979 3980 dst = xfrm_dst_child(dst); 3981 3982 if (xfrm->props.mode == XFRM_MODE_TRANSPORT) 3983 continue; 3984 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR) 3985 daddr = xfrm->coaddr; 3986 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR)) 3987 daddr = &xfrm->id.daddr; 3988 } 3989 return daddr; 3990 } 3991 3992 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst, 3993 struct sk_buff *skb, 3994 const void *daddr) 3995 { 3996 const struct dst_entry *path = xfrm_dst_path(dst); 3997 3998 if (!skb) 3999 daddr = xfrm_get_dst_nexthop(dst, daddr); 4000 return path->ops->neigh_lookup(path, skb, daddr); 4001 } 4002 4003 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr) 4004 { 4005 const struct dst_entry *path = xfrm_dst_path(dst); 4006 4007 daddr = xfrm_get_dst_nexthop(dst, daddr); 4008 path->ops->confirm_neigh(path, daddr); 4009 } 4010 4011 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family) 4012 { 4013 int err = 0; 4014 4015 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo))) 4016 return -EAFNOSUPPORT; 4017 4018 spin_lock(&xfrm_policy_afinfo_lock); 4019 if (unlikely(xfrm_policy_afinfo[family] != NULL)) 4020 err = -EEXIST; 4021 else { 4022 struct dst_ops *dst_ops = afinfo->dst_ops; 4023 if (likely(dst_ops->kmem_cachep == NULL)) 4024 dst_ops->kmem_cachep = xfrm_dst_cache; 4025 if (likely(dst_ops->check == NULL)) 4026 dst_ops->check = xfrm_dst_check; 4027 if (likely(dst_ops->default_advmss == NULL)) 4028 dst_ops->default_advmss = xfrm_default_advmss; 4029 if (likely(dst_ops->mtu == NULL)) 4030 dst_ops->mtu = xfrm_mtu; 4031 if (likely(dst_ops->negative_advice == NULL)) 4032 dst_ops->negative_advice = xfrm_negative_advice; 4033 if (likely(dst_ops->link_failure == NULL)) 4034 dst_ops->link_failure = xfrm_link_failure; 4035 if (likely(dst_ops->neigh_lookup == NULL)) 4036 dst_ops->neigh_lookup = xfrm_neigh_lookup; 4037 if (likely(!dst_ops->confirm_neigh)) 4038 dst_ops->confirm_neigh = xfrm_confirm_neigh; 4039 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo); 4040 } 4041 spin_unlock(&xfrm_policy_afinfo_lock); 4042 4043 return err; 4044 } 4045 EXPORT_SYMBOL(xfrm_policy_register_afinfo); 4046 4047 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo) 4048 { 4049 struct dst_ops *dst_ops = afinfo->dst_ops; 4050 int i; 4051 4052 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) { 4053 if (xfrm_policy_afinfo[i] != afinfo) 4054 continue; 4055 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL); 4056 break; 4057 } 4058 4059 synchronize_rcu(); 4060 4061 dst_ops->kmem_cachep = NULL; 4062 dst_ops->check = NULL; 4063 dst_ops->negative_advice = NULL; 4064 dst_ops->link_failure = NULL; 4065 } 4066 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo); 4067 4068 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb) 4069 { 4070 spin_lock(&xfrm_if_cb_lock); 4071 rcu_assign_pointer(xfrm_if_cb, ifcb); 4072 spin_unlock(&xfrm_if_cb_lock); 4073 } 4074 EXPORT_SYMBOL(xfrm_if_register_cb); 4075 4076 void xfrm_if_unregister_cb(void) 4077 { 4078 RCU_INIT_POINTER(xfrm_if_cb, NULL); 4079 synchronize_rcu(); 4080 } 4081 EXPORT_SYMBOL(xfrm_if_unregister_cb); 4082 4083 #ifdef CONFIG_XFRM_STATISTICS 4084 static int __net_init xfrm_statistics_init(struct net *net) 4085 { 4086 int rv; 4087 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib); 4088 if (!net->mib.xfrm_statistics) 4089 return -ENOMEM; 4090 rv = xfrm_proc_init(net); 4091 if (rv < 0) 4092 free_percpu(net->mib.xfrm_statistics); 4093 return rv; 4094 } 4095 4096 static void xfrm_statistics_fini(struct net *net) 4097 { 4098 xfrm_proc_fini(net); 4099 free_percpu(net->mib.xfrm_statistics); 4100 } 4101 #else 4102 static int __net_init xfrm_statistics_init(struct net *net) 4103 { 4104 return 0; 4105 } 4106 4107 static void xfrm_statistics_fini(struct net *net) 4108 { 4109 } 4110 #endif 4111 4112 static int __net_init xfrm_policy_init(struct net *net) 4113 { 4114 unsigned int hmask, sz; 4115 int dir, err; 4116 4117 if (net_eq(net, &init_net)) { 4118 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache", 4119 sizeof(struct xfrm_dst), 4120 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4121 NULL); 4122 err = rhashtable_init(&xfrm_policy_inexact_table, 4123 &xfrm_pol_inexact_params); 4124 BUG_ON(err); 4125 } 4126 4127 hmask = 8 - 1; 4128 sz = (hmask+1) * sizeof(struct hlist_head); 4129 4130 net->xfrm.policy_byidx = xfrm_hash_alloc(sz); 4131 if (!net->xfrm.policy_byidx) 4132 goto out_byidx; 4133 net->xfrm.policy_idx_hmask = hmask; 4134 4135 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 4136 struct xfrm_policy_hash *htab; 4137 4138 net->xfrm.policy_count[dir] = 0; 4139 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0; 4140 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]); 4141 4142 htab = &net->xfrm.policy_bydst[dir]; 4143 htab->table = xfrm_hash_alloc(sz); 4144 if (!htab->table) 4145 goto out_bydst; 4146 htab->hmask = hmask; 4147 htab->dbits4 = 32; 4148 htab->sbits4 = 32; 4149 htab->dbits6 = 128; 4150 htab->sbits6 = 128; 4151 } 4152 net->xfrm.policy_hthresh.lbits4 = 32; 4153 net->xfrm.policy_hthresh.rbits4 = 32; 4154 net->xfrm.policy_hthresh.lbits6 = 128; 4155 net->xfrm.policy_hthresh.rbits6 = 128; 4156 4157 seqlock_init(&net->xfrm.policy_hthresh.lock); 4158 4159 INIT_LIST_HEAD(&net->xfrm.policy_all); 4160 INIT_LIST_HEAD(&net->xfrm.inexact_bins); 4161 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize); 4162 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild); 4163 return 0; 4164 4165 out_bydst: 4166 for (dir--; dir >= 0; dir--) { 4167 struct xfrm_policy_hash *htab; 4168 4169 htab = &net->xfrm.policy_bydst[dir]; 4170 xfrm_hash_free(htab->table, sz); 4171 } 4172 xfrm_hash_free(net->xfrm.policy_byidx, sz); 4173 out_byidx: 4174 return -ENOMEM; 4175 } 4176 4177 static void xfrm_policy_fini(struct net *net) 4178 { 4179 struct xfrm_pol_inexact_bin *b, *t; 4180 unsigned int sz; 4181 int dir; 4182 4183 flush_work(&net->xfrm.policy_hash_work); 4184 #ifdef CONFIG_XFRM_SUB_POLICY 4185 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false); 4186 #endif 4187 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false); 4188 4189 WARN_ON(!list_empty(&net->xfrm.policy_all)); 4190 4191 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 4192 struct xfrm_policy_hash *htab; 4193 4194 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir])); 4195 4196 htab = &net->xfrm.policy_bydst[dir]; 4197 sz = (htab->hmask + 1) * sizeof(struct hlist_head); 4198 WARN_ON(!hlist_empty(htab->table)); 4199 xfrm_hash_free(htab->table, sz); 4200 } 4201 4202 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head); 4203 WARN_ON(!hlist_empty(net->xfrm.policy_byidx)); 4204 xfrm_hash_free(net->xfrm.policy_byidx, sz); 4205 4206 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 4207 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins) 4208 __xfrm_policy_inexact_prune_bin(b, true); 4209 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 4210 } 4211 4212 static int __net_init xfrm_net_init(struct net *net) 4213 { 4214 int rv; 4215 4216 /* Initialize the per-net locks here */ 4217 spin_lock_init(&net->xfrm.xfrm_state_lock); 4218 spin_lock_init(&net->xfrm.xfrm_policy_lock); 4219 seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock); 4220 mutex_init(&net->xfrm.xfrm_cfg_mutex); 4221 net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT; 4222 net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT; 4223 net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT; 4224 4225 rv = xfrm_statistics_init(net); 4226 if (rv < 0) 4227 goto out_statistics; 4228 rv = xfrm_state_init(net); 4229 if (rv < 0) 4230 goto out_state; 4231 rv = xfrm_policy_init(net); 4232 if (rv < 0) 4233 goto out_policy; 4234 rv = xfrm_sysctl_init(net); 4235 if (rv < 0) 4236 goto out_sysctl; 4237 4238 return 0; 4239 4240 out_sysctl: 4241 xfrm_policy_fini(net); 4242 out_policy: 4243 xfrm_state_fini(net); 4244 out_state: 4245 xfrm_statistics_fini(net); 4246 out_statistics: 4247 return rv; 4248 } 4249 4250 static void __net_exit xfrm_net_exit(struct net *net) 4251 { 4252 xfrm_sysctl_fini(net); 4253 xfrm_policy_fini(net); 4254 xfrm_state_fini(net); 4255 xfrm_statistics_fini(net); 4256 } 4257 4258 static struct pernet_operations __net_initdata xfrm_net_ops = { 4259 .init = xfrm_net_init, 4260 .exit = xfrm_net_exit, 4261 }; 4262 4263 void __init xfrm_init(void) 4264 { 4265 register_pernet_subsys(&xfrm_net_ops); 4266 xfrm_dev_init(); 4267 xfrm_input_init(); 4268 4269 #ifdef CONFIG_XFRM_ESPINTCP 4270 espintcp_init(); 4271 #endif 4272 } 4273 4274 #ifdef CONFIG_AUDITSYSCALL 4275 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp, 4276 struct audit_buffer *audit_buf) 4277 { 4278 struct xfrm_sec_ctx *ctx = xp->security; 4279 struct xfrm_selector *sel = &xp->selector; 4280 4281 if (ctx) 4282 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s", 4283 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str); 4284 4285 switch (sel->family) { 4286 case AF_INET: 4287 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4); 4288 if (sel->prefixlen_s != 32) 4289 audit_log_format(audit_buf, " src_prefixlen=%d", 4290 sel->prefixlen_s); 4291 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4); 4292 if (sel->prefixlen_d != 32) 4293 audit_log_format(audit_buf, " dst_prefixlen=%d", 4294 sel->prefixlen_d); 4295 break; 4296 case AF_INET6: 4297 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6); 4298 if (sel->prefixlen_s != 128) 4299 audit_log_format(audit_buf, " src_prefixlen=%d", 4300 sel->prefixlen_s); 4301 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6); 4302 if (sel->prefixlen_d != 128) 4303 audit_log_format(audit_buf, " dst_prefixlen=%d", 4304 sel->prefixlen_d); 4305 break; 4306 } 4307 } 4308 4309 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid) 4310 { 4311 struct audit_buffer *audit_buf; 4312 4313 audit_buf = xfrm_audit_start("SPD-add"); 4314 if (audit_buf == NULL) 4315 return; 4316 xfrm_audit_helper_usrinfo(task_valid, audit_buf); 4317 audit_log_format(audit_buf, " res=%u", result); 4318 xfrm_audit_common_policyinfo(xp, audit_buf); 4319 audit_log_end(audit_buf); 4320 } 4321 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add); 4322 4323 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result, 4324 bool task_valid) 4325 { 4326 struct audit_buffer *audit_buf; 4327 4328 audit_buf = xfrm_audit_start("SPD-delete"); 4329 if (audit_buf == NULL) 4330 return; 4331 xfrm_audit_helper_usrinfo(task_valid, audit_buf); 4332 audit_log_format(audit_buf, " res=%u", result); 4333 xfrm_audit_common_policyinfo(xp, audit_buf); 4334 audit_log_end(audit_buf); 4335 } 4336 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete); 4337 #endif 4338 4339 #ifdef CONFIG_XFRM_MIGRATE 4340 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp, 4341 const struct xfrm_selector *sel_tgt) 4342 { 4343 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) { 4344 if (sel_tgt->family == sel_cmp->family && 4345 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr, 4346 sel_cmp->family) && 4347 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr, 4348 sel_cmp->family) && 4349 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d && 4350 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) { 4351 return true; 4352 } 4353 } else { 4354 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) { 4355 return true; 4356 } 4357 } 4358 return false; 4359 } 4360 4361 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel, 4362 u8 dir, u8 type, struct net *net, u32 if_id) 4363 { 4364 struct xfrm_policy *pol, *ret = NULL; 4365 struct hlist_head *chain; 4366 u32 priority = ~0U; 4367 4368 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 4369 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir); 4370 hlist_for_each_entry(pol, chain, bydst) { 4371 if ((if_id == 0 || pol->if_id == if_id) && 4372 xfrm_migrate_selector_match(sel, &pol->selector) && 4373 pol->type == type) { 4374 ret = pol; 4375 priority = ret->priority; 4376 break; 4377 } 4378 } 4379 chain = &net->xfrm.policy_inexact[dir]; 4380 hlist_for_each_entry(pol, chain, bydst_inexact_list) { 4381 if ((pol->priority >= priority) && ret) 4382 break; 4383 4384 if ((if_id == 0 || pol->if_id == if_id) && 4385 xfrm_migrate_selector_match(sel, &pol->selector) && 4386 pol->type == type) { 4387 ret = pol; 4388 break; 4389 } 4390 } 4391 4392 xfrm_pol_hold(ret); 4393 4394 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 4395 4396 return ret; 4397 } 4398 4399 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t) 4400 { 4401 int match = 0; 4402 4403 if (t->mode == m->mode && t->id.proto == m->proto && 4404 (m->reqid == 0 || t->reqid == m->reqid)) { 4405 switch (t->mode) { 4406 case XFRM_MODE_TUNNEL: 4407 case XFRM_MODE_BEET: 4408 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr, 4409 m->old_family) && 4410 xfrm_addr_equal(&t->saddr, &m->old_saddr, 4411 m->old_family)) { 4412 match = 1; 4413 } 4414 break; 4415 case XFRM_MODE_TRANSPORT: 4416 /* in case of transport mode, template does not store 4417 any IP addresses, hence we just compare mode and 4418 protocol */ 4419 match = 1; 4420 break; 4421 default: 4422 break; 4423 } 4424 } 4425 return match; 4426 } 4427 4428 /* update endpoint address(es) of template(s) */ 4429 static int xfrm_policy_migrate(struct xfrm_policy *pol, 4430 struct xfrm_migrate *m, int num_migrate, 4431 struct netlink_ext_ack *extack) 4432 { 4433 struct xfrm_migrate *mp; 4434 int i, j, n = 0; 4435 4436 write_lock_bh(&pol->lock); 4437 if (unlikely(pol->walk.dead)) { 4438 /* target policy has been deleted */ 4439 NL_SET_ERR_MSG(extack, "Target policy not found"); 4440 write_unlock_bh(&pol->lock); 4441 return -ENOENT; 4442 } 4443 4444 for (i = 0; i < pol->xfrm_nr; i++) { 4445 for (j = 0, mp = m; j < num_migrate; j++, mp++) { 4446 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i])) 4447 continue; 4448 n++; 4449 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL && 4450 pol->xfrm_vec[i].mode != XFRM_MODE_BEET) 4451 continue; 4452 /* update endpoints */ 4453 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr, 4454 sizeof(pol->xfrm_vec[i].id.daddr)); 4455 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr, 4456 sizeof(pol->xfrm_vec[i].saddr)); 4457 pol->xfrm_vec[i].encap_family = mp->new_family; 4458 /* flush bundles */ 4459 atomic_inc(&pol->genid); 4460 } 4461 } 4462 4463 write_unlock_bh(&pol->lock); 4464 4465 if (!n) 4466 return -ENODATA; 4467 4468 return 0; 4469 } 4470 4471 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate, 4472 struct netlink_ext_ack *extack) 4473 { 4474 int i, j; 4475 4476 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH) { 4477 NL_SET_ERR_MSG(extack, "Invalid number of SAs to migrate, must be 0 < num <= XFRM_MAX_DEPTH (6)"); 4478 return -EINVAL; 4479 } 4480 4481 for (i = 0; i < num_migrate; i++) { 4482 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) || 4483 xfrm_addr_any(&m[i].new_saddr, m[i].new_family)) { 4484 NL_SET_ERR_MSG(extack, "Addresses in the MIGRATE attribute's list cannot be null"); 4485 return -EINVAL; 4486 } 4487 4488 /* check if there is any duplicated entry */ 4489 for (j = i + 1; j < num_migrate; j++) { 4490 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr, 4491 sizeof(m[i].old_daddr)) && 4492 !memcmp(&m[i].old_saddr, &m[j].old_saddr, 4493 sizeof(m[i].old_saddr)) && 4494 m[i].proto == m[j].proto && 4495 m[i].mode == m[j].mode && 4496 m[i].reqid == m[j].reqid && 4497 m[i].old_family == m[j].old_family) { 4498 NL_SET_ERR_MSG(extack, "Entries in the MIGRATE attribute's list must be unique"); 4499 return -EINVAL; 4500 } 4501 } 4502 } 4503 4504 return 0; 4505 } 4506 4507 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type, 4508 struct xfrm_migrate *m, int num_migrate, 4509 struct xfrm_kmaddress *k, struct net *net, 4510 struct xfrm_encap_tmpl *encap, u32 if_id, 4511 struct netlink_ext_ack *extack) 4512 { 4513 int i, err, nx_cur = 0, nx_new = 0; 4514 struct xfrm_policy *pol = NULL; 4515 struct xfrm_state *x, *xc; 4516 struct xfrm_state *x_cur[XFRM_MAX_DEPTH]; 4517 struct xfrm_state *x_new[XFRM_MAX_DEPTH]; 4518 struct xfrm_migrate *mp; 4519 4520 /* Stage 0 - sanity checks */ 4521 err = xfrm_migrate_check(m, num_migrate, extack); 4522 if (err < 0) 4523 goto out; 4524 4525 if (dir >= XFRM_POLICY_MAX) { 4526 NL_SET_ERR_MSG(extack, "Invalid policy direction"); 4527 err = -EINVAL; 4528 goto out; 4529 } 4530 4531 /* Stage 1 - find policy */ 4532 pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id); 4533 if (!pol) { 4534 NL_SET_ERR_MSG(extack, "Target policy not found"); 4535 err = -ENOENT; 4536 goto out; 4537 } 4538 4539 /* Stage 2 - find and update state(s) */ 4540 for (i = 0, mp = m; i < num_migrate; i++, mp++) { 4541 if ((x = xfrm_migrate_state_find(mp, net, if_id))) { 4542 x_cur[nx_cur] = x; 4543 nx_cur++; 4544 xc = xfrm_state_migrate(x, mp, encap); 4545 if (xc) { 4546 x_new[nx_new] = xc; 4547 nx_new++; 4548 } else { 4549 err = -ENODATA; 4550 goto restore_state; 4551 } 4552 } 4553 } 4554 4555 /* Stage 3 - update policy */ 4556 err = xfrm_policy_migrate(pol, m, num_migrate, extack); 4557 if (err < 0) 4558 goto restore_state; 4559 4560 /* Stage 4 - delete old state(s) */ 4561 if (nx_cur) { 4562 xfrm_states_put(x_cur, nx_cur); 4563 xfrm_states_delete(x_cur, nx_cur); 4564 } 4565 4566 /* Stage 5 - announce */ 4567 km_migrate(sel, dir, type, m, num_migrate, k, encap); 4568 4569 xfrm_pol_put(pol); 4570 4571 return 0; 4572 out: 4573 return err; 4574 4575 restore_state: 4576 if (pol) 4577 xfrm_pol_put(pol); 4578 if (nx_cur) 4579 xfrm_states_put(x_cur, nx_cur); 4580 if (nx_new) 4581 xfrm_states_delete(x_new, nx_new); 4582 4583 return err; 4584 } 4585 EXPORT_SYMBOL(xfrm_migrate); 4586 #endif 4587