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 <linux/icmp.h> 33 #include <net/dst.h> 34 #include <net/flow.h> 35 #include <net/inet_ecn.h> 36 #include <net/xfrm.h> 37 #include <net/ip.h> 38 #include <net/gre.h> 39 #if IS_ENABLED(CONFIG_IPV6_MIP6) 40 #include <net/mip6.h> 41 #endif 42 #ifdef CONFIG_XFRM_STATISTICS 43 #include <net/snmp.h> 44 #endif 45 #ifdef CONFIG_XFRM_ESPINTCP 46 #include <net/espintcp.h> 47 #endif 48 #include <net/inet_dscp.h> 49 50 #include "xfrm_hash.h" 51 52 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10)) 53 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ)) 54 #define XFRM_MAX_QUEUE_LEN 100 55 56 struct xfrm_flo { 57 struct dst_entry *dst_orig; 58 u8 flags; 59 }; 60 61 /* prefixes smaller than this are stored in lists, not trees. */ 62 #define INEXACT_PREFIXLEN_IPV4 16 63 #define INEXACT_PREFIXLEN_IPV6 48 64 65 struct xfrm_pol_inexact_node { 66 struct rb_node node; 67 union { 68 xfrm_address_t addr; 69 struct rcu_head rcu; 70 }; 71 u8 prefixlen; 72 73 struct rb_root root; 74 75 /* the policies matching this node, can be empty list */ 76 struct hlist_head hhead; 77 }; 78 79 /* xfrm inexact policy search tree: 80 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id); 81 * | 82 * +---- root_d: sorted by daddr:prefix 83 * | | 84 * | xfrm_pol_inexact_node 85 * | | 86 * | +- root: sorted by saddr/prefix 87 * | | | 88 * | | xfrm_pol_inexact_node 89 * | | | 90 * | | + root: unused 91 * | | | 92 * | | + hhead: saddr:daddr policies 93 * | | 94 * | +- coarse policies and all any:daddr policies 95 * | 96 * +---- root_s: sorted by saddr:prefix 97 * | | 98 * | xfrm_pol_inexact_node 99 * | | 100 * | + root: unused 101 * | | 102 * | + hhead: saddr:any policies 103 * | 104 * +---- coarse policies and all any:any policies 105 * 106 * Lookups return four candidate lists: 107 * 1. any:any list from top-level xfrm_pol_inexact_bin 108 * 2. any:daddr list from daddr tree 109 * 3. saddr:daddr list from 2nd level daddr tree 110 * 4. saddr:any list from saddr tree 111 * 112 * This result set then needs to be searched for the policy with 113 * the lowest priority. If two candidates have the same priority, the 114 * struct xfrm_policy pos member with the lower number is used. 115 * 116 * This replicates previous single-list-search algorithm which would 117 * return first matching policy in the (ordered-by-priority) list. 118 */ 119 120 struct xfrm_pol_inexact_key { 121 possible_net_t net; 122 u32 if_id; 123 u16 family; 124 u8 dir, type; 125 }; 126 127 struct xfrm_pol_inexact_bin { 128 struct xfrm_pol_inexact_key k; 129 struct rhash_head head; 130 /* list containing '*:*' policies */ 131 struct hlist_head hhead; 132 133 seqcount_spinlock_t count; 134 /* tree sorted by daddr/prefix */ 135 struct rb_root root_d; 136 137 /* tree sorted by saddr/prefix */ 138 struct rb_root root_s; 139 140 /* slow path below */ 141 struct list_head inexact_bins; 142 struct rcu_head rcu; 143 }; 144 145 enum xfrm_pol_inexact_candidate_type { 146 XFRM_POL_CAND_BOTH, 147 XFRM_POL_CAND_SADDR, 148 XFRM_POL_CAND_DADDR, 149 XFRM_POL_CAND_ANY, 150 151 XFRM_POL_CAND_MAX, 152 }; 153 154 struct xfrm_pol_inexact_candidates { 155 struct hlist_head *res[XFRM_POL_CAND_MAX]; 156 }; 157 158 struct xfrm_flow_keys { 159 struct flow_dissector_key_basic basic; 160 struct flow_dissector_key_control control; 161 union { 162 struct flow_dissector_key_ipv4_addrs ipv4; 163 struct flow_dissector_key_ipv6_addrs ipv6; 164 } addrs; 165 struct flow_dissector_key_ip ip; 166 struct flow_dissector_key_icmp icmp; 167 struct flow_dissector_key_ports ports; 168 struct flow_dissector_key_keyid gre; 169 }; 170 171 static struct flow_dissector xfrm_session_dissector __ro_after_init; 172 173 static DEFINE_SPINLOCK(xfrm_if_cb_lock); 174 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly; 175 176 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock); 177 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1] 178 __read_mostly; 179 180 static struct kmem_cache *xfrm_dst_cache __ro_after_init; 181 182 static struct rhashtable xfrm_policy_inexact_table; 183 static const struct rhashtable_params xfrm_pol_inexact_params; 184 185 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr); 186 static int stale_bundle(struct dst_entry *dst); 187 static int xfrm_bundle_ok(struct xfrm_dst *xdst); 188 static void xfrm_policy_queue_process(struct timer_list *t); 189 190 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir); 191 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol, 192 int dir); 193 194 static struct xfrm_pol_inexact_bin * 195 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir, 196 u32 if_id); 197 198 static struct xfrm_pol_inexact_bin * 199 xfrm_policy_inexact_lookup_rcu(struct net *net, 200 u8 type, u16 family, u8 dir, u32 if_id); 201 static struct xfrm_policy * 202 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy, 203 bool excl); 204 205 static bool 206 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand, 207 struct xfrm_pol_inexact_bin *b, 208 const xfrm_address_t *saddr, 209 const xfrm_address_t *daddr); 210 211 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy) 212 { 213 return refcount_inc_not_zero(&policy->refcnt); 214 } 215 216 static inline bool 217 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl) 218 { 219 const struct flowi4 *fl4 = &fl->u.ip4; 220 221 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) && 222 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) && 223 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) && 224 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) && 225 (fl4->flowi4_proto == sel->proto || !sel->proto) && 226 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex); 227 } 228 229 static inline bool 230 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl) 231 { 232 const struct flowi6 *fl6 = &fl->u.ip6; 233 234 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) && 235 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) && 236 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) && 237 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) && 238 (fl6->flowi6_proto == sel->proto || !sel->proto) && 239 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex); 240 } 241 242 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl, 243 unsigned short family) 244 { 245 if (family != sel->family && sel->family != AF_UNSPEC) 246 return false; 247 248 switch (family) { 249 case AF_INET: 250 return __xfrm4_selector_match(sel, fl); 251 case AF_INET6: 252 return __xfrm6_selector_match(sel, fl); 253 } 254 return false; 255 } 256 257 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family) 258 { 259 const struct xfrm_policy_afinfo *afinfo; 260 261 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo))) 262 return NULL; 263 rcu_read_lock(); 264 afinfo = rcu_dereference(xfrm_policy_afinfo[family]); 265 if (unlikely(!afinfo)) 266 rcu_read_unlock(); 267 return afinfo; 268 } 269 270 /* Called with rcu_read_lock(). */ 271 static const struct xfrm_if_cb *xfrm_if_get_cb(void) 272 { 273 return rcu_dereference(xfrm_if_cb); 274 } 275 276 struct dst_entry *__xfrm_dst_lookup(int family, 277 const struct xfrm_dst_lookup_params *params) 278 { 279 const struct xfrm_policy_afinfo *afinfo; 280 struct dst_entry *dst; 281 282 afinfo = xfrm_policy_get_afinfo(family); 283 if (unlikely(afinfo == NULL)) 284 return ERR_PTR(-EAFNOSUPPORT); 285 286 dst = afinfo->dst_lookup(params); 287 288 rcu_read_unlock(); 289 290 return dst; 291 } 292 EXPORT_SYMBOL(__xfrm_dst_lookup); 293 294 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, 295 dscp_t dscp, int oif, 296 xfrm_address_t *prev_saddr, 297 xfrm_address_t *prev_daddr, 298 int family, u32 mark) 299 { 300 struct xfrm_dst_lookup_params params; 301 struct net *net = xs_net(x); 302 xfrm_address_t *saddr = &x->props.saddr; 303 xfrm_address_t *daddr = &x->id.daddr; 304 struct dst_entry *dst; 305 306 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) { 307 saddr = x->coaddr; 308 daddr = prev_daddr; 309 } 310 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) { 311 saddr = prev_saddr; 312 daddr = x->coaddr; 313 } 314 315 params.net = net; 316 params.saddr = saddr; 317 params.daddr = daddr; 318 params.dscp = dscp; 319 params.oif = oif; 320 params.mark = mark; 321 params.ipproto = x->id.proto; 322 if (x->encap) { 323 switch (x->encap->encap_type) { 324 case UDP_ENCAP_ESPINUDP: 325 params.ipproto = IPPROTO_UDP; 326 params.uli.ports.sport = x->encap->encap_sport; 327 params.uli.ports.dport = x->encap->encap_dport; 328 break; 329 case TCP_ENCAP_ESPINTCP: 330 params.ipproto = IPPROTO_TCP; 331 params.uli.ports.sport = x->encap->encap_sport; 332 params.uli.ports.dport = x->encap->encap_dport; 333 break; 334 } 335 } 336 337 dst = __xfrm_dst_lookup(family, ¶ms); 338 339 if (!IS_ERR(dst)) { 340 if (prev_saddr != saddr) 341 memcpy(prev_saddr, saddr, sizeof(*prev_saddr)); 342 if (prev_daddr != daddr) 343 memcpy(prev_daddr, daddr, sizeof(*prev_daddr)); 344 } 345 346 return dst; 347 } 348 349 static inline unsigned long make_jiffies(long secs) 350 { 351 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ) 352 return MAX_SCHEDULE_TIMEOUT-1; 353 else 354 return secs*HZ; 355 } 356 357 static void xfrm_policy_timer(struct timer_list *t) 358 { 359 struct xfrm_policy *xp = timer_container_of(xp, t, timer); 360 time64_t now = ktime_get_real_seconds(); 361 time64_t next = TIME64_MAX; 362 int warn = 0; 363 int dir; 364 365 read_lock(&xp->lock); 366 367 if (unlikely(xp->walk.dead)) 368 goto out; 369 370 dir = xfrm_policy_id2dir(xp->index); 371 372 if (xp->lft.hard_add_expires_seconds) { 373 time64_t tmo = xp->lft.hard_add_expires_seconds + 374 xp->curlft.add_time - now; 375 if (tmo <= 0) 376 goto expired; 377 if (tmo < next) 378 next = tmo; 379 } 380 if (xp->lft.hard_use_expires_seconds) { 381 time64_t tmo = xp->lft.hard_use_expires_seconds + 382 (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now; 383 if (tmo <= 0) 384 goto expired; 385 if (tmo < next) 386 next = tmo; 387 } 388 if (xp->lft.soft_add_expires_seconds) { 389 time64_t tmo = xp->lft.soft_add_expires_seconds + 390 xp->curlft.add_time - now; 391 if (tmo <= 0) { 392 warn = 1; 393 tmo = XFRM_KM_TIMEOUT; 394 } 395 if (tmo < next) 396 next = tmo; 397 } 398 if (xp->lft.soft_use_expires_seconds) { 399 time64_t tmo = xp->lft.soft_use_expires_seconds + 400 (READ_ONCE(xp->curlft.use_time) ? : xp->curlft.add_time) - now; 401 if (tmo <= 0) { 402 warn = 1; 403 tmo = XFRM_KM_TIMEOUT; 404 } 405 if (tmo < next) 406 next = tmo; 407 } 408 409 if (warn) 410 km_policy_expired(xp, dir, 0, 0); 411 if (next != TIME64_MAX && 412 !mod_timer(&xp->timer, jiffies + make_jiffies(next))) 413 xfrm_pol_hold(xp); 414 415 out: 416 read_unlock(&xp->lock); 417 xfrm_pol_put(xp); 418 return; 419 420 expired: 421 read_unlock(&xp->lock); 422 if (!xfrm_policy_delete(xp, dir)) 423 km_policy_expired(xp, dir, 1, 0); 424 xfrm_pol_put(xp); 425 } 426 427 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2 428 * SPD calls. 429 */ 430 431 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp) 432 { 433 struct xfrm_policy *policy; 434 435 policy = kzalloc_obj(struct xfrm_policy, gfp); 436 437 if (policy) { 438 write_pnet(&policy->xp_net, net); 439 INIT_LIST_HEAD(&policy->walk.all); 440 INIT_HLIST_HEAD(&policy->state_cache_list); 441 INIT_HLIST_NODE(&policy->bydst); 442 INIT_HLIST_NODE(&policy->byidx); 443 rwlock_init(&policy->lock); 444 refcount_set(&policy->refcnt, 1); 445 skb_queue_head_init(&policy->polq.hold_queue); 446 timer_setup(&policy->timer, xfrm_policy_timer, 0); 447 timer_setup(&policy->polq.hold_timer, 448 xfrm_policy_queue_process, 0); 449 } 450 return policy; 451 } 452 EXPORT_SYMBOL(xfrm_policy_alloc); 453 454 static void xfrm_policy_destroy_rcu(struct rcu_head *head) 455 { 456 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu); 457 458 security_xfrm_policy_free(policy->security); 459 kfree(policy); 460 } 461 462 /* Destroy xfrm_policy: descendant resources must be released to this moment. */ 463 464 void xfrm_policy_destroy(struct xfrm_policy *policy) 465 { 466 BUG_ON(!policy->walk.dead); 467 468 if (timer_delete(&policy->timer) || timer_delete(&policy->polq.hold_timer)) 469 BUG(); 470 471 xfrm_dev_policy_free(policy); 472 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu); 473 } 474 EXPORT_SYMBOL(xfrm_policy_destroy); 475 476 /* Rule must be locked. Release descendant resources, announce 477 * entry dead. The rule must be unlinked from lists to the moment. 478 */ 479 480 static void xfrm_policy_kill(struct xfrm_policy *policy) 481 { 482 struct net *net = xp_net(policy); 483 struct xfrm_state *x; 484 485 xfrm_dev_policy_delete(policy); 486 487 write_lock_bh(&policy->lock); 488 policy->walk.dead = 1; 489 write_unlock_bh(&policy->lock); 490 491 atomic_inc(&policy->genid); 492 493 if (timer_delete(&policy->polq.hold_timer)) 494 xfrm_pol_put(policy); 495 skb_queue_purge(&policy->polq.hold_queue); 496 497 if (timer_delete(&policy->timer)) 498 xfrm_pol_put(policy); 499 500 /* XXX: Flush state cache */ 501 spin_lock_bh(&net->xfrm.xfrm_state_lock); 502 hlist_for_each_entry_rcu(x, &policy->state_cache_list, state_cache) { 503 hlist_del_init_rcu(&x->state_cache); 504 } 505 spin_unlock_bh(&net->xfrm.xfrm_state_lock); 506 507 xfrm_pol_put(policy); 508 } 509 510 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024; 511 512 static inline unsigned int idx_hash(struct net *net, u32 index) 513 { 514 return __idx_hash(index, net->xfrm.policy_idx_hmask); 515 } 516 517 /* calculate policy hash thresholds */ 518 static void __get_hash_thresh(struct net *net, 519 unsigned short family, int dir, 520 u8 *dbits, u8 *sbits) 521 { 522 switch (family) { 523 case AF_INET: 524 *dbits = net->xfrm.policy_bydst[dir].dbits4; 525 *sbits = net->xfrm.policy_bydst[dir].sbits4; 526 break; 527 528 case AF_INET6: 529 *dbits = net->xfrm.policy_bydst[dir].dbits6; 530 *sbits = net->xfrm.policy_bydst[dir].sbits6; 531 break; 532 533 default: 534 *dbits = 0; 535 *sbits = 0; 536 } 537 } 538 539 static struct hlist_head *policy_hash_bysel(struct net *net, 540 const struct xfrm_selector *sel, 541 unsigned short family, int dir) 542 { 543 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 544 unsigned int hash; 545 u8 dbits; 546 u8 sbits; 547 548 __get_hash_thresh(net, family, dir, &dbits, &sbits); 549 hash = __sel_hash(sel, family, hmask, dbits, sbits); 550 551 if (hash == hmask + 1) 552 return NULL; 553 554 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table, 555 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash; 556 } 557 558 static struct hlist_head *policy_hash_direct(struct net *net, 559 const xfrm_address_t *daddr, 560 const xfrm_address_t *saddr, 561 unsigned short family, int dir) 562 { 563 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 564 unsigned int hash; 565 u8 dbits; 566 u8 sbits; 567 568 __get_hash_thresh(net, family, dir, &dbits, &sbits); 569 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits); 570 571 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table, 572 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash; 573 } 574 575 static void xfrm_dst_hash_transfer(struct net *net, 576 struct hlist_head *list, 577 struct hlist_head *ndsttable, 578 unsigned int nhashmask, 579 int dir) 580 { 581 struct hlist_node *tmp, *entry0 = NULL; 582 struct xfrm_policy *pol; 583 unsigned int h0 = 0; 584 u8 dbits; 585 u8 sbits; 586 587 redo: 588 hlist_for_each_entry_safe(pol, tmp, list, bydst) { 589 unsigned int h; 590 591 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits); 592 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr, 593 pol->family, nhashmask, dbits, sbits); 594 if (!entry0 || pol->xdo.type == XFRM_DEV_OFFLOAD_PACKET) { 595 hlist_del_rcu(&pol->bydst); 596 hlist_add_head_rcu(&pol->bydst, ndsttable + h); 597 h0 = h; 598 } else { 599 if (h != h0) 600 continue; 601 hlist_del_rcu(&pol->bydst); 602 hlist_add_behind_rcu(&pol->bydst, entry0); 603 } 604 entry0 = &pol->bydst; 605 } 606 if (!hlist_empty(list)) { 607 entry0 = NULL; 608 goto redo; 609 } 610 } 611 612 static void xfrm_idx_hash_transfer(struct hlist_head *list, 613 struct hlist_head *nidxtable, 614 unsigned int nhashmask) 615 { 616 struct hlist_node *tmp; 617 struct xfrm_policy *pol; 618 619 hlist_for_each_entry_safe(pol, tmp, list, byidx) { 620 unsigned int h; 621 622 h = __idx_hash(pol->index, nhashmask); 623 hlist_add_head(&pol->byidx, nidxtable+h); 624 } 625 } 626 627 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask) 628 { 629 return ((old_hmask + 1) << 1) - 1; 630 } 631 632 static void xfrm_bydst_resize(struct net *net, int dir) 633 { 634 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 635 unsigned int nhashmask = xfrm_new_hash_mask(hmask); 636 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head); 637 struct hlist_head *ndst = xfrm_hash_alloc(nsize); 638 struct hlist_head *odst; 639 int i; 640 641 if (!ndst) 642 return; 643 644 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 645 write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation); 646 647 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table, 648 lockdep_is_held(&net->xfrm.xfrm_policy_lock)); 649 650 for (i = hmask; i >= 0; i--) 651 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir); 652 653 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst); 654 net->xfrm.policy_bydst[dir].hmask = nhashmask; 655 656 write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation); 657 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 658 659 synchronize_rcu(); 660 661 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head)); 662 } 663 664 static void xfrm_byidx_resize(struct net *net) 665 { 666 unsigned int hmask = net->xfrm.policy_idx_hmask; 667 unsigned int nhashmask = xfrm_new_hash_mask(hmask); 668 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head); 669 struct hlist_head *oidx = net->xfrm.policy_byidx; 670 struct hlist_head *nidx = xfrm_hash_alloc(nsize); 671 int i; 672 673 if (!nidx) 674 return; 675 676 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 677 678 for (i = hmask; i >= 0; i--) 679 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask); 680 681 net->xfrm.policy_byidx = nidx; 682 net->xfrm.policy_idx_hmask = nhashmask; 683 684 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 685 686 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head)); 687 } 688 689 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total) 690 { 691 unsigned int cnt = READ_ONCE(net->xfrm.policy_count[dir]); 692 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 693 694 if (total) 695 *total += cnt; 696 697 if ((hmask + 1) < xfrm_policy_hashmax && 698 cnt > hmask) 699 return 1; 700 701 return 0; 702 } 703 704 static inline int xfrm_byidx_should_resize(struct net *net, int total) 705 { 706 unsigned int hmask = net->xfrm.policy_idx_hmask; 707 708 if ((hmask + 1) < xfrm_policy_hashmax && 709 total > hmask) 710 return 1; 711 712 return 0; 713 } 714 715 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si) 716 { 717 si->incnt = READ_ONCE(net->xfrm.policy_count[XFRM_POLICY_IN]); 718 si->outcnt = READ_ONCE(net->xfrm.policy_count[XFRM_POLICY_OUT]); 719 si->fwdcnt = READ_ONCE(net->xfrm.policy_count[XFRM_POLICY_FWD]); 720 si->inscnt = READ_ONCE(net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX]); 721 si->outscnt = READ_ONCE(net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX]); 722 si->fwdscnt = READ_ONCE(net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX]); 723 si->spdhcnt = net->xfrm.policy_idx_hmask; 724 si->spdhmcnt = xfrm_policy_hashmax; 725 } 726 EXPORT_SYMBOL(xfrm_spd_getinfo); 727 728 static DEFINE_MUTEX(hash_resize_mutex); 729 static void xfrm_hash_resize(struct work_struct *work) 730 { 731 struct net *net = container_of(work, struct net, xfrm.policy_hash_work); 732 int dir, total; 733 734 mutex_lock(&hash_resize_mutex); 735 736 total = 0; 737 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 738 if (xfrm_bydst_should_resize(net, dir, &total)) 739 xfrm_bydst_resize(net, dir); 740 } 741 if (xfrm_byidx_should_resize(net, total)) 742 xfrm_byidx_resize(net); 743 744 mutex_unlock(&hash_resize_mutex); 745 } 746 747 /* Make sure *pol can be inserted into fastbin. 748 * Useful to check that later insert requests will be successful 749 * (provided xfrm_policy_lock is held throughout). 750 */ 751 static struct xfrm_pol_inexact_bin * 752 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir) 753 { 754 struct xfrm_pol_inexact_bin *bin, *prev; 755 struct xfrm_pol_inexact_key k = { 756 .family = pol->family, 757 .type = pol->type, 758 .dir = dir, 759 .if_id = pol->if_id, 760 }; 761 struct net *net = xp_net(pol); 762 763 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 764 765 write_pnet(&k.net, net); 766 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k, 767 xfrm_pol_inexact_params); 768 if (bin) 769 return bin; 770 771 bin = kzalloc_obj(*bin, GFP_ATOMIC); 772 if (!bin) 773 return NULL; 774 775 bin->k = k; 776 INIT_HLIST_HEAD(&bin->hhead); 777 bin->root_d = RB_ROOT; 778 bin->root_s = RB_ROOT; 779 seqcount_spinlock_init(&bin->count, &net->xfrm.xfrm_policy_lock); 780 781 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table, 782 &bin->k, &bin->head, 783 xfrm_pol_inexact_params); 784 if (!prev) { 785 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins); 786 return bin; 787 } 788 789 kfree(bin); 790 791 return IS_ERR(prev) ? NULL : prev; 792 } 793 794 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr, 795 int family, u8 prefixlen) 796 { 797 if (xfrm_addr_any(addr, family)) 798 return true; 799 800 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6) 801 return true; 802 803 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4) 804 return true; 805 806 return false; 807 } 808 809 static bool 810 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy) 811 { 812 const xfrm_address_t *addr; 813 bool saddr_any, daddr_any; 814 u8 prefixlen; 815 816 addr = &policy->selector.saddr; 817 prefixlen = policy->selector.prefixlen_s; 818 819 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr, 820 policy->family, 821 prefixlen); 822 addr = &policy->selector.daddr; 823 prefixlen = policy->selector.prefixlen_d; 824 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr, 825 policy->family, 826 prefixlen); 827 return saddr_any && daddr_any; 828 } 829 830 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node, 831 const xfrm_address_t *addr, u8 prefixlen) 832 { 833 node->addr = *addr; 834 node->prefixlen = prefixlen; 835 } 836 837 static struct xfrm_pol_inexact_node * 838 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen) 839 { 840 struct xfrm_pol_inexact_node *node; 841 842 node = kzalloc_obj(*node, GFP_ATOMIC); 843 if (node) 844 xfrm_pol_inexact_node_init(node, addr, prefixlen); 845 846 return node; 847 } 848 849 static int xfrm_policy_addr_delta(const xfrm_address_t *a, 850 const xfrm_address_t *b, 851 u8 prefixlen, u16 family) 852 { 853 u32 ma, mb, mask; 854 unsigned int pdw, pbi; 855 int delta = 0; 856 857 switch (family) { 858 case AF_INET: 859 if (prefixlen == 0) 860 return 0; 861 mask = ~0U << (32 - prefixlen); 862 ma = ntohl(a->a4) & mask; 863 mb = ntohl(b->a4) & mask; 864 if (ma < mb) 865 delta = -1; 866 else if (ma > mb) 867 delta = 1; 868 break; 869 case AF_INET6: 870 pdw = prefixlen >> 5; 871 pbi = prefixlen & 0x1f; 872 873 if (pdw) { 874 delta = memcmp(a->a6, b->a6, pdw << 2); 875 if (delta) 876 return delta; 877 } 878 if (pbi) { 879 mask = ~0U << (32 - pbi); 880 ma = ntohl(a->a6[pdw]) & mask; 881 mb = ntohl(b->a6[pdw]) & mask; 882 if (ma < mb) 883 delta = -1; 884 else if (ma > mb) 885 delta = 1; 886 } 887 break; 888 default: 889 break; 890 } 891 892 return delta; 893 } 894 895 static void xfrm_policy_inexact_list_reinsert(struct net *net, 896 struct xfrm_pol_inexact_node *n, 897 u16 family) 898 { 899 unsigned int matched_s, matched_d; 900 struct xfrm_policy *policy, *p; 901 902 matched_s = 0; 903 matched_d = 0; 904 905 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) { 906 struct hlist_node *newpos = NULL; 907 bool matches_s, matches_d; 908 909 if (policy->walk.dead || !policy->bydst_reinsert) 910 continue; 911 912 WARN_ON_ONCE(policy->family != family); 913 914 policy->bydst_reinsert = false; 915 hlist_for_each_entry(p, &n->hhead, bydst) { 916 if (policy->priority > p->priority) 917 newpos = &p->bydst; 918 else if (policy->priority == p->priority && 919 policy->pos > p->pos) 920 newpos = &p->bydst; 921 else 922 break; 923 } 924 925 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET) 926 hlist_add_behind_rcu(&policy->bydst, newpos); 927 else 928 hlist_add_head_rcu(&policy->bydst, &n->hhead); 929 930 /* paranoia checks follow. 931 * Check that the reinserted policy matches at least 932 * saddr or daddr for current node prefix. 933 * 934 * Matching both is fine, matching saddr in one policy 935 * (but not daddr) and then matching only daddr in another 936 * is a bug. 937 */ 938 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr, 939 &n->addr, 940 n->prefixlen, 941 family) == 0; 942 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr, 943 &n->addr, 944 n->prefixlen, 945 family) == 0; 946 if (matches_s && matches_d) 947 continue; 948 949 WARN_ON_ONCE(!matches_s && !matches_d); 950 if (matches_s) 951 matched_s++; 952 if (matches_d) 953 matched_d++; 954 WARN_ON_ONCE(matched_s && matched_d); 955 } 956 } 957 958 static void xfrm_policy_inexact_node_reinsert(struct net *net, 959 struct xfrm_pol_inexact_node *n, 960 struct rb_root *new, 961 u16 family) 962 { 963 struct xfrm_pol_inexact_node *node; 964 struct rb_node **p, *parent; 965 966 /* we should not have another subtree here */ 967 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root)); 968 restart: 969 parent = NULL; 970 p = &new->rb_node; 971 while (*p) { 972 u8 prefixlen; 973 int delta; 974 975 parent = *p; 976 node = rb_entry(*p, struct xfrm_pol_inexact_node, node); 977 978 prefixlen = min(node->prefixlen, n->prefixlen); 979 980 delta = xfrm_policy_addr_delta(&n->addr, &node->addr, 981 prefixlen, family); 982 if (delta < 0) { 983 p = &parent->rb_left; 984 } else if (delta > 0) { 985 p = &parent->rb_right; 986 } else { 987 bool same_prefixlen = node->prefixlen == n->prefixlen; 988 struct xfrm_policy *tmp; 989 990 hlist_for_each_entry(tmp, &n->hhead, bydst) { 991 tmp->bydst_reinsert = true; 992 hlist_del_rcu(&tmp->bydst); 993 } 994 995 node->prefixlen = prefixlen; 996 997 xfrm_policy_inexact_list_reinsert(net, node, family); 998 999 if (same_prefixlen) { 1000 kfree_rcu(n, rcu); 1001 return; 1002 } 1003 1004 rb_erase(*p, new); 1005 kfree_rcu(n, rcu); 1006 n = node; 1007 goto restart; 1008 } 1009 } 1010 1011 rb_link_node_rcu(&n->node, parent, p); 1012 rb_insert_color(&n->node, new); 1013 } 1014 1015 /* merge nodes v and n */ 1016 static void xfrm_policy_inexact_node_merge(struct net *net, 1017 struct xfrm_pol_inexact_node *v, 1018 struct xfrm_pol_inexact_node *n, 1019 u16 family) 1020 { 1021 struct xfrm_pol_inexact_node *node; 1022 struct xfrm_policy *tmp; 1023 struct rb_node *rnode; 1024 1025 /* To-be-merged node v has a subtree. 1026 * 1027 * Dismantle it and insert its nodes to n->root. 1028 */ 1029 while ((rnode = rb_first(&v->root)) != NULL) { 1030 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node); 1031 rb_erase(&node->node, &v->root); 1032 xfrm_policy_inexact_node_reinsert(net, node, &n->root, 1033 family); 1034 } 1035 1036 hlist_for_each_entry(tmp, &v->hhead, bydst) { 1037 tmp->bydst_reinsert = true; 1038 hlist_del_rcu(&tmp->bydst); 1039 } 1040 1041 xfrm_policy_inexact_list_reinsert(net, n, family); 1042 } 1043 1044 static struct xfrm_pol_inexact_node * 1045 xfrm_policy_inexact_insert_node(struct net *net, 1046 struct rb_root *root, 1047 xfrm_address_t *addr, 1048 u16 family, u8 prefixlen, u8 dir) 1049 { 1050 struct xfrm_pol_inexact_node *cached = NULL; 1051 struct rb_node **p, *parent = NULL; 1052 struct xfrm_pol_inexact_node *node; 1053 1054 p = &root->rb_node; 1055 while (*p) { 1056 int delta; 1057 1058 parent = *p; 1059 node = rb_entry(*p, struct xfrm_pol_inexact_node, node); 1060 1061 delta = xfrm_policy_addr_delta(addr, &node->addr, 1062 node->prefixlen, 1063 family); 1064 if (delta == 0 && prefixlen >= node->prefixlen) { 1065 WARN_ON_ONCE(cached); /* ipsec policies got lost */ 1066 return node; 1067 } 1068 1069 if (delta < 0) 1070 p = &parent->rb_left; 1071 else 1072 p = &parent->rb_right; 1073 1074 if (prefixlen < node->prefixlen) { 1075 delta = xfrm_policy_addr_delta(addr, &node->addr, 1076 prefixlen, 1077 family); 1078 if (delta) 1079 continue; 1080 1081 /* This node is a subnet of the new prefix. It needs 1082 * to be removed and re-inserted with the smaller 1083 * prefix and all nodes that are now also covered 1084 * by the reduced prefixlen. 1085 */ 1086 rb_erase(&node->node, root); 1087 1088 if (!cached) { 1089 xfrm_pol_inexact_node_init(node, addr, 1090 prefixlen); 1091 cached = node; 1092 } else { 1093 /* This node also falls within the new 1094 * prefixlen. Merge the to-be-reinserted 1095 * node and this one. 1096 */ 1097 xfrm_policy_inexact_node_merge(net, node, 1098 cached, family); 1099 kfree_rcu(node, rcu); 1100 } 1101 1102 /* restart */ 1103 p = &root->rb_node; 1104 parent = NULL; 1105 } 1106 } 1107 1108 node = cached; 1109 if (!node) { 1110 node = xfrm_pol_inexact_node_alloc(addr, prefixlen); 1111 if (!node) 1112 return NULL; 1113 } 1114 1115 rb_link_node_rcu(&node->node, parent, p); 1116 rb_insert_color(&node->node, root); 1117 1118 return node; 1119 } 1120 1121 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm) 1122 { 1123 struct xfrm_pol_inexact_node *node; 1124 struct rb_node *rn = rb_first(r); 1125 1126 while (rn) { 1127 node = rb_entry(rn, struct xfrm_pol_inexact_node, node); 1128 1129 xfrm_policy_inexact_gc_tree(&node->root, rm); 1130 rn = rb_next(rn); 1131 1132 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) { 1133 WARN_ON_ONCE(rm); 1134 continue; 1135 } 1136 1137 rb_erase(&node->node, r); 1138 kfree_rcu(node, rcu); 1139 } 1140 } 1141 1142 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit) 1143 { 1144 write_seqcount_begin(&b->count); 1145 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit); 1146 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit); 1147 write_seqcount_end(&b->count); 1148 1149 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) || 1150 !hlist_empty(&b->hhead)) { 1151 WARN_ON_ONCE(net_exit); 1152 return; 1153 } 1154 1155 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head, 1156 xfrm_pol_inexact_params) == 0) { 1157 list_del(&b->inexact_bins); 1158 kfree_rcu(b, rcu); 1159 } 1160 } 1161 1162 static void __xfrm_policy_inexact_flush(struct net *net) 1163 { 1164 struct xfrm_pol_inexact_bin *bin, *t; 1165 1166 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1167 1168 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins) 1169 __xfrm_policy_inexact_prune_bin(bin, false); 1170 } 1171 1172 static struct hlist_head * 1173 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin, 1174 struct xfrm_policy *policy, u8 dir) 1175 { 1176 struct xfrm_pol_inexact_node *n; 1177 struct net *net; 1178 1179 net = xp_net(policy); 1180 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1181 1182 if (xfrm_policy_inexact_insert_use_any_list(policy)) 1183 return &bin->hhead; 1184 1185 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr, 1186 policy->family, 1187 policy->selector.prefixlen_d)) { 1188 write_seqcount_begin(&bin->count); 1189 n = xfrm_policy_inexact_insert_node(net, 1190 &bin->root_s, 1191 &policy->selector.saddr, 1192 policy->family, 1193 policy->selector.prefixlen_s, 1194 dir); 1195 write_seqcount_end(&bin->count); 1196 if (!n) 1197 return NULL; 1198 1199 return &n->hhead; 1200 } 1201 1202 /* daddr is fixed */ 1203 write_seqcount_begin(&bin->count); 1204 n = xfrm_policy_inexact_insert_node(net, 1205 &bin->root_d, 1206 &policy->selector.daddr, 1207 policy->family, 1208 policy->selector.prefixlen_d, dir); 1209 write_seqcount_end(&bin->count); 1210 if (!n) 1211 return NULL; 1212 1213 /* saddr is wildcard */ 1214 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr, 1215 policy->family, 1216 policy->selector.prefixlen_s)) 1217 return &n->hhead; 1218 1219 write_seqcount_begin(&bin->count); 1220 n = xfrm_policy_inexact_insert_node(net, 1221 &n->root, 1222 &policy->selector.saddr, 1223 policy->family, 1224 policy->selector.prefixlen_s, dir); 1225 write_seqcount_end(&bin->count); 1226 if (!n) 1227 return NULL; 1228 1229 return &n->hhead; 1230 } 1231 1232 static struct xfrm_policy * 1233 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl) 1234 { 1235 struct xfrm_pol_inexact_bin *bin; 1236 struct xfrm_policy *delpol; 1237 struct hlist_head *chain; 1238 struct net *net; 1239 1240 bin = xfrm_policy_inexact_alloc_bin(policy, dir); 1241 if (!bin) 1242 return ERR_PTR(-ENOMEM); 1243 1244 net = xp_net(policy); 1245 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1246 1247 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir); 1248 if (!chain) { 1249 __xfrm_policy_inexact_prune_bin(bin, false); 1250 return ERR_PTR(-ENOMEM); 1251 } 1252 1253 delpol = xfrm_policy_insert_list(chain, policy, excl); 1254 if (delpol && excl) { 1255 __xfrm_policy_inexact_prune_bin(bin, false); 1256 return ERR_PTR(-EEXIST); 1257 } 1258 1259 if (delpol) 1260 __xfrm_policy_inexact_prune_bin(bin, false); 1261 1262 return delpol; 1263 } 1264 1265 static bool xfrm_policy_is_dead_or_sk(const struct xfrm_policy *policy) 1266 { 1267 int dir; 1268 1269 if (policy->walk.dead) 1270 return true; 1271 1272 dir = xfrm_policy_id2dir(policy->index); 1273 return dir >= XFRM_POLICY_MAX; 1274 } 1275 1276 static void xfrm_hash_rebuild(struct work_struct *work) 1277 { 1278 struct net *net = container_of(work, struct net, 1279 xfrm.policy_hthresh.work); 1280 struct xfrm_policy *pol; 1281 struct xfrm_policy *policy; 1282 struct hlist_head *chain; 1283 struct hlist_node *newpos; 1284 int dir; 1285 unsigned seq; 1286 u8 lbits4, rbits4, lbits6, rbits6; 1287 1288 mutex_lock(&hash_resize_mutex); 1289 1290 /* read selector prefixlen thresholds */ 1291 do { 1292 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock); 1293 1294 lbits4 = net->xfrm.policy_hthresh.lbits4; 1295 rbits4 = net->xfrm.policy_hthresh.rbits4; 1296 lbits6 = net->xfrm.policy_hthresh.lbits6; 1297 rbits6 = net->xfrm.policy_hthresh.rbits6; 1298 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq)); 1299 1300 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1301 write_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation); 1302 1303 /* make sure that we can insert the indirect policies again before 1304 * we start with destructive action. 1305 */ 1306 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) { 1307 struct xfrm_pol_inexact_bin *bin; 1308 u8 dbits, sbits; 1309 1310 if (xfrm_policy_is_dead_or_sk(policy)) 1311 continue; 1312 1313 dir = xfrm_policy_id2dir(policy->index); 1314 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) { 1315 if (policy->family == AF_INET) { 1316 dbits = rbits4; 1317 sbits = lbits4; 1318 } else { 1319 dbits = rbits6; 1320 sbits = lbits6; 1321 } 1322 } else { 1323 if (policy->family == AF_INET) { 1324 dbits = lbits4; 1325 sbits = rbits4; 1326 } else { 1327 dbits = lbits6; 1328 sbits = rbits6; 1329 } 1330 } 1331 1332 if (policy->selector.prefixlen_d < dbits || 1333 policy->selector.prefixlen_s < sbits) 1334 continue; 1335 1336 bin = xfrm_policy_inexact_alloc_bin(policy, dir); 1337 if (!bin) 1338 goto out_unlock; 1339 1340 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir)) 1341 goto out_unlock; 1342 } 1343 1344 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 1345 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) { 1346 /* dir out => dst = remote, src = local */ 1347 net->xfrm.policy_bydst[dir].dbits4 = rbits4; 1348 net->xfrm.policy_bydst[dir].sbits4 = lbits4; 1349 net->xfrm.policy_bydst[dir].dbits6 = rbits6; 1350 net->xfrm.policy_bydst[dir].sbits6 = lbits6; 1351 } else { 1352 /* dir in/fwd => dst = local, src = remote */ 1353 net->xfrm.policy_bydst[dir].dbits4 = lbits4; 1354 net->xfrm.policy_bydst[dir].sbits4 = rbits4; 1355 net->xfrm.policy_bydst[dir].dbits6 = lbits6; 1356 net->xfrm.policy_bydst[dir].sbits6 = rbits6; 1357 } 1358 } 1359 1360 /* re-insert all policies by order of creation */ 1361 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) { 1362 if (xfrm_policy_is_dead_or_sk(policy)) 1363 continue; 1364 1365 hlist_del_rcu(&policy->bydst); 1366 1367 newpos = NULL; 1368 dir = xfrm_policy_id2dir(policy->index); 1369 chain = policy_hash_bysel(net, &policy->selector, 1370 policy->family, dir); 1371 1372 if (!chain) { 1373 void *p = xfrm_policy_inexact_insert(policy, dir, 0); 1374 1375 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p)); 1376 continue; 1377 } 1378 1379 hlist_for_each_entry(pol, chain, bydst) { 1380 if (policy->priority >= pol->priority) 1381 newpos = &pol->bydst; 1382 else 1383 break; 1384 } 1385 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET) 1386 hlist_add_behind_rcu(&policy->bydst, newpos); 1387 else 1388 hlist_add_head_rcu(&policy->bydst, chain); 1389 } 1390 1391 out_unlock: 1392 __xfrm_policy_inexact_flush(net); 1393 write_seqcount_end(&net->xfrm.xfrm_policy_hash_generation); 1394 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1395 1396 mutex_unlock(&hash_resize_mutex); 1397 } 1398 1399 void xfrm_policy_hash_rebuild(struct net *net) 1400 { 1401 schedule_work(&net->xfrm.policy_hthresh.work); 1402 } 1403 EXPORT_SYMBOL(xfrm_policy_hash_rebuild); 1404 1405 /* Generate new index... KAME seems to generate them ordered by cost 1406 * of an absolute inpredictability of ordering of rules. This will not pass. */ 1407 static u32 xfrm_gen_index(struct net *net, int dir, u32 index) 1408 { 1409 for (;;) { 1410 struct hlist_head *list; 1411 struct xfrm_policy *p; 1412 u32 idx; 1413 int found; 1414 1415 if (!index) { 1416 idx = (net->xfrm.idx_generator | dir); 1417 net->xfrm.idx_generator += 8; 1418 } else { 1419 idx = index; 1420 index = 0; 1421 } 1422 1423 if (idx == 0) 1424 idx = 8; 1425 list = net->xfrm.policy_byidx + idx_hash(net, idx); 1426 found = 0; 1427 hlist_for_each_entry(p, list, byidx) { 1428 if (p->index == idx) { 1429 found = 1; 1430 break; 1431 } 1432 } 1433 if (!found) 1434 return idx; 1435 } 1436 } 1437 1438 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2) 1439 { 1440 u32 *p1 = (u32 *) s1; 1441 u32 *p2 = (u32 *) s2; 1442 int len = sizeof(struct xfrm_selector) / sizeof(u32); 1443 int i; 1444 1445 for (i = 0; i < len; i++) { 1446 if (p1[i] != p2[i]) 1447 return 1; 1448 } 1449 1450 return 0; 1451 } 1452 1453 static void xfrm_policy_requeue(struct xfrm_policy *old, 1454 struct xfrm_policy *new) 1455 { 1456 struct xfrm_policy_queue *pq = &old->polq; 1457 struct sk_buff_head list; 1458 1459 if (skb_queue_empty(&pq->hold_queue)) 1460 return; 1461 1462 __skb_queue_head_init(&list); 1463 1464 spin_lock_bh(&pq->hold_queue.lock); 1465 skb_queue_splice_init(&pq->hold_queue, &list); 1466 if (timer_delete(&pq->hold_timer)) 1467 xfrm_pol_put(old); 1468 spin_unlock_bh(&pq->hold_queue.lock); 1469 1470 pq = &new->polq; 1471 1472 spin_lock_bh(&pq->hold_queue.lock); 1473 skb_queue_splice(&list, &pq->hold_queue); 1474 pq->timeout = XFRM_QUEUE_TMO_MIN; 1475 if (!mod_timer(&pq->hold_timer, jiffies)) 1476 xfrm_pol_hold(new); 1477 spin_unlock_bh(&pq->hold_queue.lock); 1478 } 1479 1480 static inline bool xfrm_policy_mark_match(const struct xfrm_mark *mark, 1481 struct xfrm_policy *pol) 1482 { 1483 return mark->v == pol->mark.v && mark->m == pol->mark.m; 1484 } 1485 1486 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed) 1487 { 1488 const struct xfrm_pol_inexact_key *k = data; 1489 u32 a = k->type << 24 | k->dir << 16 | k->family; 1490 1491 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)), 1492 seed); 1493 } 1494 1495 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed) 1496 { 1497 const struct xfrm_pol_inexact_bin *b = data; 1498 1499 return xfrm_pol_bin_key(&b->k, 0, seed); 1500 } 1501 1502 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg, 1503 const void *ptr) 1504 { 1505 const struct xfrm_pol_inexact_key *key = arg->key; 1506 const struct xfrm_pol_inexact_bin *b = ptr; 1507 int ret; 1508 1509 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net))) 1510 return -1; 1511 1512 ret = b->k.dir ^ key->dir; 1513 if (ret) 1514 return ret; 1515 1516 ret = b->k.type ^ key->type; 1517 if (ret) 1518 return ret; 1519 1520 ret = b->k.family ^ key->family; 1521 if (ret) 1522 return ret; 1523 1524 return b->k.if_id ^ key->if_id; 1525 } 1526 1527 static const struct rhashtable_params xfrm_pol_inexact_params = { 1528 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head), 1529 .hashfn = xfrm_pol_bin_key, 1530 .obj_hashfn = xfrm_pol_bin_obj, 1531 .obj_cmpfn = xfrm_pol_bin_cmp, 1532 .automatic_shrinking = true, 1533 }; 1534 1535 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain, 1536 struct xfrm_policy *policy, 1537 bool excl) 1538 { 1539 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL; 1540 1541 hlist_for_each_entry(pol, chain, bydst) { 1542 if (pol->type == policy->type && 1543 pol->if_id == policy->if_id && 1544 !selector_cmp(&pol->selector, &policy->selector) && 1545 xfrm_policy_mark_match(&policy->mark, pol) && 1546 xfrm_sec_ctx_match(pol->security, policy->security) && 1547 !WARN_ON(delpol)) { 1548 if (excl) 1549 return ERR_PTR(-EEXIST); 1550 delpol = pol; 1551 if (policy->priority > pol->priority) 1552 continue; 1553 } else if (policy->priority >= pol->priority) { 1554 newpos = pol; 1555 continue; 1556 } 1557 if (delpol) 1558 break; 1559 } 1560 1561 if (newpos && policy->xdo.type != XFRM_DEV_OFFLOAD_PACKET) 1562 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst); 1563 else 1564 /* Packet offload policies enter to the head 1565 * to speed-up lookups. 1566 */ 1567 hlist_add_head_rcu(&policy->bydst, chain); 1568 1569 return delpol; 1570 } 1571 1572 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl) 1573 { 1574 struct net *net = xp_net(policy); 1575 struct xfrm_policy *delpol; 1576 struct hlist_head *chain; 1577 1578 /* Sanitize mark before store */ 1579 policy->mark.v &= policy->mark.m; 1580 1581 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1582 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir); 1583 if (chain) 1584 delpol = xfrm_policy_insert_list(chain, policy, excl); 1585 else 1586 delpol = xfrm_policy_inexact_insert(policy, dir, excl); 1587 1588 if (IS_ERR(delpol)) { 1589 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1590 return PTR_ERR(delpol); 1591 } 1592 1593 __xfrm_policy_link(policy, dir); 1594 1595 /* After previous checking, family can either be AF_INET or AF_INET6 */ 1596 if (policy->family == AF_INET) 1597 rt_genid_bump_ipv4(net); 1598 else 1599 rt_genid_bump_ipv6(net); 1600 1601 if (delpol) { 1602 xfrm_policy_requeue(delpol, policy); 1603 __xfrm_policy_unlink(delpol, dir); 1604 } 1605 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index); 1606 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index)); 1607 policy->curlft.add_time = ktime_get_real_seconds(); 1608 policy->curlft.use_time = 0; 1609 if (!mod_timer(&policy->timer, jiffies + HZ)) 1610 xfrm_pol_hold(policy); 1611 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1612 1613 if (delpol) 1614 xfrm_policy_kill(delpol); 1615 else if (xfrm_bydst_should_resize(net, dir, NULL)) 1616 schedule_work(&net->xfrm.policy_hash_work); 1617 1618 return 0; 1619 } 1620 EXPORT_SYMBOL(xfrm_policy_insert); 1621 1622 static struct xfrm_policy * 1623 __xfrm_policy_bysel_ctx(struct hlist_head *chain, const struct xfrm_mark *mark, 1624 u32 if_id, u8 type, int dir, struct xfrm_selector *sel, 1625 struct xfrm_sec_ctx *ctx) 1626 { 1627 struct xfrm_policy *pol; 1628 1629 if (!chain) 1630 return NULL; 1631 1632 hlist_for_each_entry(pol, chain, bydst) { 1633 if (pol->type == type && 1634 pol->if_id == if_id && 1635 xfrm_policy_mark_match(mark, pol) && 1636 !selector_cmp(sel, &pol->selector) && 1637 xfrm_sec_ctx_match(ctx, pol->security)) 1638 return pol; 1639 } 1640 1641 return NULL; 1642 } 1643 1644 struct xfrm_policy * 1645 xfrm_policy_bysel_ctx(struct net *net, const struct xfrm_mark *mark, u32 if_id, 1646 u8 type, int dir, struct xfrm_selector *sel, 1647 struct xfrm_sec_ctx *ctx, int delete, int *err) 1648 { 1649 struct xfrm_pol_inexact_bin *bin = NULL; 1650 struct xfrm_policy *pol, *ret = NULL; 1651 struct hlist_head *chain; 1652 1653 *err = 0; 1654 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1655 chain = policy_hash_bysel(net, sel, sel->family, dir); 1656 if (!chain) { 1657 struct xfrm_pol_inexact_candidates cand; 1658 int i; 1659 1660 bin = xfrm_policy_inexact_lookup(net, type, 1661 sel->family, dir, if_id); 1662 if (!bin) { 1663 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1664 return NULL; 1665 } 1666 1667 if (!xfrm_policy_find_inexact_candidates(&cand, bin, 1668 &sel->saddr, 1669 &sel->daddr)) { 1670 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1671 return NULL; 1672 } 1673 1674 pol = NULL; 1675 for (i = 0; i < ARRAY_SIZE(cand.res); i++) { 1676 struct xfrm_policy *tmp; 1677 1678 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark, 1679 if_id, type, dir, 1680 sel, ctx); 1681 if (!tmp) 1682 continue; 1683 1684 if (!pol || tmp->pos < pol->pos) 1685 pol = tmp; 1686 } 1687 } else { 1688 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir, 1689 sel, ctx); 1690 } 1691 1692 if (pol) { 1693 xfrm_pol_hold(pol); 1694 if (delete) { 1695 *err = security_xfrm_policy_delete(pol->security); 1696 if (*err) { 1697 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1698 return pol; 1699 } 1700 __xfrm_policy_unlink(pol, dir); 1701 } 1702 ret = pol; 1703 } 1704 if (bin && delete) 1705 __xfrm_policy_inexact_prune_bin(bin, false); 1706 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1707 1708 if (ret && delete) 1709 xfrm_policy_kill(ret); 1710 return ret; 1711 } 1712 EXPORT_SYMBOL(xfrm_policy_bysel_ctx); 1713 1714 struct xfrm_policy * 1715 xfrm_policy_byid(struct net *net, const struct xfrm_mark *mark, u32 if_id, 1716 u8 type, int dir, u32 id, int delete, int *err) 1717 { 1718 struct xfrm_policy *pol, *ret; 1719 struct hlist_head *chain; 1720 1721 *err = -ENOENT; 1722 if (xfrm_policy_id2dir(id) != dir) 1723 return NULL; 1724 1725 *err = 0; 1726 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1727 chain = net->xfrm.policy_byidx + idx_hash(net, id); 1728 ret = NULL; 1729 hlist_for_each_entry(pol, chain, byidx) { 1730 if (pol->type == type && pol->index == id && 1731 pol->if_id == if_id && xfrm_policy_mark_match(mark, pol)) { 1732 xfrm_pol_hold(pol); 1733 if (delete) { 1734 *err = security_xfrm_policy_delete( 1735 pol->security); 1736 if (*err) { 1737 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1738 return pol; 1739 } 1740 __xfrm_policy_unlink(pol, dir); 1741 } 1742 ret = pol; 1743 break; 1744 } 1745 } 1746 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1747 1748 if (ret && delete) 1749 xfrm_policy_kill(ret); 1750 return ret; 1751 } 1752 EXPORT_SYMBOL(xfrm_policy_byid); 1753 1754 #ifdef CONFIG_SECURITY_NETWORK_XFRM 1755 static inline int 1756 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid) 1757 { 1758 struct xfrm_policy *pol; 1759 int err = 0; 1760 1761 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1762 if (pol->walk.dead || 1763 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX || 1764 pol->type != type) 1765 continue; 1766 1767 err = security_xfrm_policy_delete(pol->security); 1768 if (err) { 1769 xfrm_audit_policy_delete(pol, 0, task_valid); 1770 return err; 1771 } 1772 } 1773 return err; 1774 } 1775 1776 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net, 1777 struct net_device *dev, 1778 bool task_valid) 1779 { 1780 struct xfrm_policy *pol; 1781 int err = 0; 1782 1783 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1784 if (pol->walk.dead || 1785 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX || 1786 pol->xdo.dev != dev) 1787 continue; 1788 1789 err = security_xfrm_policy_delete(pol->security); 1790 if (err) { 1791 xfrm_audit_policy_delete(pol, 0, task_valid); 1792 return err; 1793 } 1794 } 1795 return err; 1796 } 1797 #else 1798 static inline int 1799 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid) 1800 { 1801 return 0; 1802 } 1803 1804 static inline int xfrm_dev_policy_flush_secctx_check(struct net *net, 1805 struct net_device *dev, 1806 bool task_valid) 1807 { 1808 return 0; 1809 } 1810 #endif 1811 1812 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid) 1813 { 1814 int dir, err = 0, cnt = 0; 1815 struct xfrm_policy *pol; 1816 1817 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1818 1819 err = xfrm_policy_flush_secctx_check(net, type, task_valid); 1820 if (err) 1821 goto out; 1822 1823 again: 1824 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1825 if (pol->walk.dead) 1826 continue; 1827 1828 dir = xfrm_policy_id2dir(pol->index); 1829 if (dir >= XFRM_POLICY_MAX || 1830 pol->type != type) 1831 continue; 1832 1833 __xfrm_policy_unlink(pol, dir); 1834 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1835 cnt++; 1836 xfrm_audit_policy_delete(pol, 1, task_valid); 1837 xfrm_policy_kill(pol); 1838 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1839 goto again; 1840 } 1841 if (cnt) 1842 __xfrm_policy_inexact_flush(net); 1843 else 1844 err = -ESRCH; 1845 out: 1846 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1847 return err; 1848 } 1849 EXPORT_SYMBOL(xfrm_policy_flush); 1850 1851 int xfrm_dev_policy_flush(struct net *net, struct net_device *dev, 1852 bool task_valid) 1853 { 1854 int dir, err = 0, cnt = 0; 1855 struct xfrm_policy *pol; 1856 1857 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1858 1859 err = xfrm_dev_policy_flush_secctx_check(net, dev, task_valid); 1860 if (err) 1861 goto out; 1862 1863 again: 1864 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1865 if (pol->walk.dead) 1866 continue; 1867 1868 dir = xfrm_policy_id2dir(pol->index); 1869 if (dir >= XFRM_POLICY_MAX || 1870 pol->xdo.dev != dev) 1871 continue; 1872 1873 __xfrm_policy_unlink(pol, dir); 1874 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1875 cnt++; 1876 xfrm_audit_policy_delete(pol, 1, task_valid); 1877 xfrm_policy_kill(pol); 1878 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1879 goto again; 1880 } 1881 if (cnt) 1882 __xfrm_policy_inexact_flush(net); 1883 else 1884 err = -ESRCH; 1885 out: 1886 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1887 return err; 1888 } 1889 EXPORT_SYMBOL(xfrm_dev_policy_flush); 1890 1891 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk, 1892 int (*func)(struct xfrm_policy *, int, int, void*), 1893 void *data) 1894 { 1895 struct xfrm_policy *pol; 1896 struct xfrm_policy_walk_entry *x; 1897 int error = 0; 1898 1899 if (walk->type >= XFRM_POLICY_TYPE_MAX && 1900 walk->type != XFRM_POLICY_TYPE_ANY) 1901 return -EINVAL; 1902 1903 if (list_empty(&walk->walk.all) && walk->seq != 0) 1904 return 0; 1905 1906 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1907 if (list_empty(&walk->walk.all)) 1908 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all); 1909 else 1910 x = list_first_entry(&walk->walk.all, 1911 struct xfrm_policy_walk_entry, all); 1912 1913 list_for_each_entry_from(x, &net->xfrm.policy_all, all) { 1914 if (x->dead) 1915 continue; 1916 pol = container_of(x, struct xfrm_policy, walk); 1917 if (walk->type != XFRM_POLICY_TYPE_ANY && 1918 walk->type != pol->type) 1919 continue; 1920 error = func(pol, xfrm_policy_id2dir(pol->index), 1921 walk->seq, data); 1922 if (error) { 1923 list_move_tail(&walk->walk.all, &x->all); 1924 goto out; 1925 } 1926 walk->seq++; 1927 } 1928 if (walk->seq == 0) { 1929 error = -ENOENT; 1930 goto out; 1931 } 1932 list_del_init(&walk->walk.all); 1933 out: 1934 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1935 return error; 1936 } 1937 EXPORT_SYMBOL(xfrm_policy_walk); 1938 1939 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type) 1940 { 1941 INIT_LIST_HEAD(&walk->walk.all); 1942 walk->walk.dead = 1; 1943 walk->type = type; 1944 walk->seq = 0; 1945 } 1946 EXPORT_SYMBOL(xfrm_policy_walk_init); 1947 1948 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net) 1949 { 1950 if (list_empty(&walk->walk.all)) 1951 return; 1952 1953 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */ 1954 list_del(&walk->walk.all); 1955 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1956 } 1957 EXPORT_SYMBOL(xfrm_policy_walk_done); 1958 1959 /* 1960 * Find policy to apply to this flow. 1961 * 1962 * Returns 0 if policy found, else an -errno. 1963 */ 1964 static int xfrm_policy_match(const struct xfrm_policy *pol, 1965 const struct flowi *fl, 1966 u8 type, u16 family, u32 if_id) 1967 { 1968 const struct xfrm_selector *sel = &pol->selector; 1969 int ret = -ESRCH; 1970 bool match; 1971 1972 if (pol->family != family || 1973 pol->if_id != if_id || 1974 (fl->flowi_mark & pol->mark.m) != pol->mark.v || 1975 pol->type != type) 1976 return ret; 1977 1978 match = xfrm_selector_match(sel, fl, family); 1979 if (match) 1980 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid); 1981 return ret; 1982 } 1983 1984 static struct xfrm_pol_inexact_node * 1985 xfrm_policy_lookup_inexact_addr(const struct rb_root *r, 1986 seqcount_spinlock_t *count, 1987 const xfrm_address_t *addr, u16 family) 1988 { 1989 const struct rb_node *parent; 1990 int seq; 1991 1992 again: 1993 seq = read_seqcount_begin(count); 1994 1995 parent = rcu_dereference_raw(r->rb_node); 1996 while (parent) { 1997 struct xfrm_pol_inexact_node *node; 1998 int delta; 1999 2000 node = rb_entry(parent, struct xfrm_pol_inexact_node, node); 2001 2002 delta = xfrm_policy_addr_delta(addr, &node->addr, 2003 node->prefixlen, family); 2004 if (delta < 0) { 2005 parent = rcu_dereference_raw(parent->rb_left); 2006 continue; 2007 } else if (delta > 0) { 2008 parent = rcu_dereference_raw(parent->rb_right); 2009 continue; 2010 } 2011 2012 return node; 2013 } 2014 2015 if (read_seqcount_retry(count, seq)) 2016 goto again; 2017 2018 return NULL; 2019 } 2020 2021 static bool 2022 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand, 2023 struct xfrm_pol_inexact_bin *b, 2024 const xfrm_address_t *saddr, 2025 const xfrm_address_t *daddr) 2026 { 2027 struct xfrm_pol_inexact_node *n; 2028 u16 family; 2029 2030 if (!b) 2031 return false; 2032 2033 family = b->k.family; 2034 memset(cand, 0, sizeof(*cand)); 2035 cand->res[XFRM_POL_CAND_ANY] = &b->hhead; 2036 2037 n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr, 2038 family); 2039 if (n) { 2040 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead; 2041 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr, 2042 family); 2043 if (n) 2044 cand->res[XFRM_POL_CAND_BOTH] = &n->hhead; 2045 } 2046 2047 n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr, 2048 family); 2049 if (n) 2050 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead; 2051 2052 return true; 2053 } 2054 2055 static struct xfrm_pol_inexact_bin * 2056 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family, 2057 u8 dir, u32 if_id) 2058 { 2059 struct xfrm_pol_inexact_key k = { 2060 .family = family, 2061 .type = type, 2062 .dir = dir, 2063 .if_id = if_id, 2064 }; 2065 2066 write_pnet(&k.net, net); 2067 2068 return rhashtable_lookup(&xfrm_policy_inexact_table, &k, 2069 xfrm_pol_inexact_params); 2070 } 2071 2072 static struct xfrm_pol_inexact_bin * 2073 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, 2074 u8 dir, u32 if_id) 2075 { 2076 struct xfrm_pol_inexact_bin *bin; 2077 2078 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 2079 2080 rcu_read_lock(); 2081 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id); 2082 rcu_read_unlock(); 2083 2084 return bin; 2085 } 2086 2087 static struct xfrm_policy * 2088 __xfrm_policy_eval_candidates(struct hlist_head *chain, 2089 struct xfrm_policy *prefer, 2090 const struct flowi *fl, 2091 u8 type, u16 family, u32 if_id) 2092 { 2093 u32 priority = prefer ? prefer->priority : ~0u; 2094 struct xfrm_policy *pol; 2095 2096 if (!chain) 2097 return NULL; 2098 2099 hlist_for_each_entry_rcu(pol, chain, bydst) { 2100 int err; 2101 2102 if (pol->priority > priority) 2103 break; 2104 2105 err = xfrm_policy_match(pol, fl, type, family, if_id); 2106 if (err) { 2107 if (err != -ESRCH) 2108 return ERR_PTR(err); 2109 2110 continue; 2111 } 2112 2113 if (prefer) { 2114 /* matches. Is it older than *prefer? */ 2115 if (pol->priority == priority && 2116 prefer->pos < pol->pos) 2117 return prefer; 2118 } 2119 2120 return pol; 2121 } 2122 2123 return NULL; 2124 } 2125 2126 static struct xfrm_policy * 2127 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand, 2128 struct xfrm_policy *prefer, 2129 const struct flowi *fl, 2130 u8 type, u16 family, u32 if_id) 2131 { 2132 struct xfrm_policy *tmp; 2133 int i; 2134 2135 for (i = 0; i < ARRAY_SIZE(cand->res); i++) { 2136 tmp = __xfrm_policy_eval_candidates(cand->res[i], 2137 prefer, 2138 fl, type, family, if_id); 2139 if (!tmp) 2140 continue; 2141 2142 if (IS_ERR(tmp)) 2143 return tmp; 2144 prefer = tmp; 2145 } 2146 2147 return prefer; 2148 } 2149 2150 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type, 2151 const struct flowi *fl, 2152 u16 family, u8 dir, 2153 u32 if_id) 2154 { 2155 struct xfrm_pol_inexact_candidates cand; 2156 const xfrm_address_t *daddr, *saddr; 2157 struct xfrm_pol_inexact_bin *bin; 2158 struct xfrm_policy *pol, *ret; 2159 struct hlist_head *chain; 2160 unsigned int sequence; 2161 int err; 2162 2163 daddr = xfrm_flowi_daddr(fl, family); 2164 saddr = xfrm_flowi_saddr(fl, family); 2165 if (unlikely(!daddr || !saddr)) 2166 return NULL; 2167 2168 rcu_read_lock(); 2169 retry: 2170 do { 2171 sequence = read_seqcount_begin(&net->xfrm.xfrm_policy_hash_generation); 2172 chain = policy_hash_direct(net, daddr, saddr, family, dir); 2173 } while (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence)); 2174 2175 ret = NULL; 2176 hlist_for_each_entry_rcu(pol, chain, bydst) { 2177 err = xfrm_policy_match(pol, fl, type, family, if_id); 2178 if (err) { 2179 if (err == -ESRCH) 2180 continue; 2181 else { 2182 ret = ERR_PTR(err); 2183 goto fail; 2184 } 2185 } else { 2186 ret = pol; 2187 break; 2188 } 2189 } 2190 if (ret && ret->xdo.type == XFRM_DEV_OFFLOAD_PACKET) 2191 goto skip_inexact; 2192 2193 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id); 2194 if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr, 2195 daddr)) 2196 goto skip_inexact; 2197 2198 pol = xfrm_policy_eval_candidates(&cand, ret, fl, type, 2199 family, if_id); 2200 if (pol) { 2201 ret = pol; 2202 if (IS_ERR(pol)) 2203 goto fail; 2204 } 2205 2206 skip_inexact: 2207 if (read_seqcount_retry(&net->xfrm.xfrm_policy_hash_generation, sequence)) 2208 goto retry; 2209 2210 if (ret && !xfrm_pol_hold_rcu(ret)) 2211 goto retry; 2212 fail: 2213 rcu_read_unlock(); 2214 2215 return ret; 2216 } 2217 2218 static struct xfrm_policy *xfrm_policy_lookup(struct net *net, 2219 const struct flowi *fl, 2220 u16 family, u8 dir, u32 if_id) 2221 { 2222 #ifdef CONFIG_XFRM_SUB_POLICY 2223 struct xfrm_policy *pol; 2224 2225 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, 2226 dir, if_id); 2227 if (pol != NULL) 2228 return pol; 2229 #endif 2230 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, 2231 dir, if_id); 2232 } 2233 2234 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir, 2235 const struct flowi *fl, 2236 u16 family, u32 if_id) 2237 { 2238 struct xfrm_policy *pol; 2239 2240 rcu_read_lock(); 2241 again: 2242 pol = rcu_dereference(sk->sk_policy[dir]); 2243 if (pol != NULL) { 2244 bool match; 2245 int err = 0; 2246 2247 if (pol->family != family) { 2248 pol = NULL; 2249 goto out; 2250 } 2251 2252 match = xfrm_selector_match(&pol->selector, fl, family); 2253 if (match) { 2254 if ((READ_ONCE(sk->sk_mark) & pol->mark.m) != pol->mark.v || 2255 pol->if_id != if_id) { 2256 pol = NULL; 2257 goto out; 2258 } 2259 err = security_xfrm_policy_lookup(pol->security, 2260 fl->flowi_secid); 2261 if (!err) { 2262 if (!xfrm_pol_hold_rcu(pol)) 2263 goto again; 2264 } else if (err == -ESRCH) { 2265 pol = NULL; 2266 } else { 2267 pol = ERR_PTR(err); 2268 } 2269 } else 2270 pol = NULL; 2271 } 2272 out: 2273 rcu_read_unlock(); 2274 return pol; 2275 } 2276 2277 static u32 xfrm_gen_pos_slow(struct net *net) 2278 { 2279 struct xfrm_policy *policy; 2280 u32 i = 0; 2281 2282 /* oldest entry is last in list */ 2283 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) { 2284 if (!xfrm_policy_is_dead_or_sk(policy)) 2285 policy->pos = ++i; 2286 } 2287 2288 return i; 2289 } 2290 2291 static u32 xfrm_gen_pos(struct net *net) 2292 { 2293 const struct xfrm_policy *policy; 2294 u32 i = 0; 2295 2296 /* most recently added policy is at the head of the list */ 2297 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) { 2298 if (xfrm_policy_is_dead_or_sk(policy)) 2299 continue; 2300 2301 if (policy->pos == UINT_MAX) 2302 return xfrm_gen_pos_slow(net); 2303 2304 i = policy->pos + 1; 2305 break; 2306 } 2307 2308 return i; 2309 } 2310 2311 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir) 2312 { 2313 struct net *net = xp_net(pol); 2314 2315 switch (dir) { 2316 case XFRM_POLICY_IN: 2317 case XFRM_POLICY_FWD: 2318 case XFRM_POLICY_OUT: 2319 pol->pos = xfrm_gen_pos(net); 2320 break; 2321 } 2322 2323 list_add(&pol->walk.all, &net->xfrm.policy_all); 2324 WRITE_ONCE(net->xfrm.policy_count[dir], net->xfrm.policy_count[dir] + 1); 2325 xfrm_pol_hold(pol); 2326 } 2327 2328 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol, 2329 int dir) 2330 { 2331 struct net *net = xp_net(pol); 2332 2333 if (list_empty(&pol->walk.all)) 2334 return NULL; 2335 2336 /* Socket policies are not hashed. */ 2337 if (!hlist_unhashed(&pol->bydst)) { 2338 hlist_del_rcu(&pol->bydst); 2339 hlist_del(&pol->byidx); 2340 } 2341 2342 list_del_init(&pol->walk.all); 2343 WRITE_ONCE(net->xfrm.policy_count[dir], net->xfrm.policy_count[dir] - 1); 2344 2345 return pol; 2346 } 2347 2348 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir) 2349 { 2350 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir); 2351 } 2352 2353 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir) 2354 { 2355 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir); 2356 } 2357 2358 int xfrm_policy_delete(struct xfrm_policy *pol, int dir) 2359 { 2360 struct net *net = xp_net(pol); 2361 2362 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 2363 pol = __xfrm_policy_unlink(pol, dir); 2364 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 2365 if (pol) { 2366 xfrm_policy_kill(pol); 2367 return 0; 2368 } 2369 return -ENOENT; 2370 } 2371 EXPORT_SYMBOL(xfrm_policy_delete); 2372 2373 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol) 2374 { 2375 struct net *net = sock_net(sk); 2376 struct xfrm_policy *old_pol; 2377 2378 #ifdef CONFIG_XFRM_SUB_POLICY 2379 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN) 2380 return -EINVAL; 2381 #endif 2382 2383 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 2384 old_pol = rcu_dereference_protected(sk->sk_policy[dir], 2385 lockdep_is_held(&net->xfrm.xfrm_policy_lock)); 2386 if (pol) { 2387 pol->curlft.add_time = ktime_get_real_seconds(); 2388 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0); 2389 xfrm_sk_policy_link(pol, dir); 2390 } 2391 rcu_assign_pointer(sk->sk_policy[dir], pol); 2392 if (old_pol) { 2393 if (pol) 2394 xfrm_policy_requeue(old_pol, pol); 2395 2396 /* Unlinking succeeds always. This is the only function 2397 * allowed to delete or replace socket policy. 2398 */ 2399 xfrm_sk_policy_unlink(old_pol, dir); 2400 } 2401 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 2402 2403 if (old_pol) { 2404 xfrm_policy_kill(old_pol); 2405 } 2406 return 0; 2407 } 2408 2409 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir) 2410 { 2411 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC); 2412 struct net *net = xp_net(old); 2413 2414 if (newp) { 2415 newp->selector = old->selector; 2416 if (security_xfrm_policy_clone(old->security, 2417 &newp->security)) { 2418 kfree(newp); 2419 return NULL; /* ENOMEM */ 2420 } 2421 newp->lft = old->lft; 2422 newp->curlft = old->curlft; 2423 newp->mark = old->mark; 2424 newp->if_id = old->if_id; 2425 newp->action = old->action; 2426 newp->flags = old->flags; 2427 newp->xfrm_nr = old->xfrm_nr; 2428 newp->index = old->index; 2429 newp->type = old->type; 2430 newp->family = old->family; 2431 memcpy(newp->xfrm_vec, old->xfrm_vec, 2432 newp->xfrm_nr*sizeof(struct xfrm_tmpl)); 2433 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 2434 xfrm_sk_policy_link(newp, dir); 2435 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 2436 xfrm_pol_put(newp); 2437 } 2438 return newp; 2439 } 2440 2441 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk) 2442 { 2443 const struct xfrm_policy *p; 2444 struct xfrm_policy *np; 2445 int i, ret = 0; 2446 2447 rcu_read_lock(); 2448 for (i = 0; i < 2; i++) { 2449 p = rcu_dereference(osk->sk_policy[i]); 2450 if (p) { 2451 np = clone_policy(p, i); 2452 if (unlikely(!np)) { 2453 ret = -ENOMEM; 2454 break; 2455 } 2456 rcu_assign_pointer(sk->sk_policy[i], np); 2457 } 2458 } 2459 rcu_read_unlock(); 2460 return ret; 2461 } 2462 2463 static int 2464 xfrm_get_saddr(unsigned short family, xfrm_address_t *saddr, 2465 const struct xfrm_dst_lookup_params *params) 2466 { 2467 int err; 2468 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 2469 2470 if (unlikely(afinfo == NULL)) 2471 return -EINVAL; 2472 err = afinfo->get_saddr(saddr, params); 2473 rcu_read_unlock(); 2474 return err; 2475 } 2476 2477 /* Resolve list of templates for the flow, given policy. */ 2478 2479 static int 2480 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl, 2481 struct xfrm_state **xfrm, unsigned short family) 2482 { 2483 struct net *net = xp_net(policy); 2484 int nx; 2485 int i, error; 2486 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family); 2487 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family); 2488 xfrm_address_t tmp; 2489 2490 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) { 2491 struct xfrm_state *x; 2492 xfrm_address_t *remote = daddr; 2493 xfrm_address_t *local = saddr; 2494 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i]; 2495 2496 if (tmpl->mode == XFRM_MODE_TUNNEL || 2497 tmpl->mode == XFRM_MODE_IPTFS || 2498 tmpl->mode == XFRM_MODE_BEET) { 2499 remote = &tmpl->id.daddr; 2500 local = &tmpl->saddr; 2501 if (xfrm_addr_any(local, tmpl->encap_family)) { 2502 struct xfrm_dst_lookup_params params; 2503 2504 memset(¶ms, 0, sizeof(params)); 2505 params.net = net; 2506 params.oif = fl->flowi_oif; 2507 params.daddr = remote; 2508 error = xfrm_get_saddr(tmpl->encap_family, &tmp, 2509 ¶ms); 2510 if (error) 2511 goto fail; 2512 local = &tmp; 2513 } 2514 } 2515 2516 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, 2517 family, policy->if_id); 2518 if (x && x->dir && x->dir != XFRM_SA_DIR_OUT) { 2519 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEDIRERROR); 2520 xfrm_state_put(x); 2521 error = -EINVAL; 2522 goto fail; 2523 } 2524 2525 if (x && x->km.state == XFRM_STATE_VALID) { 2526 xfrm[nx++] = x; 2527 daddr = remote; 2528 saddr = local; 2529 continue; 2530 } 2531 if (x) { 2532 error = (x->km.state == XFRM_STATE_ERROR ? 2533 -EINVAL : -EAGAIN); 2534 xfrm_state_put(x); 2535 } else if (error == -ESRCH) { 2536 error = -EAGAIN; 2537 } 2538 2539 if (!tmpl->optional) 2540 goto fail; 2541 } 2542 return nx; 2543 2544 fail: 2545 for (nx--; nx >= 0; nx--) 2546 xfrm_state_put(xfrm[nx]); 2547 return error; 2548 } 2549 2550 static int 2551 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl, 2552 struct xfrm_state **xfrm, unsigned short family) 2553 { 2554 struct xfrm_state *tp[XFRM_MAX_DEPTH]; 2555 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm; 2556 int cnx = 0; 2557 int error; 2558 int ret; 2559 int i; 2560 2561 for (i = 0; i < npols; i++) { 2562 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) { 2563 error = -ENOBUFS; 2564 goto fail; 2565 } 2566 2567 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family); 2568 if (ret < 0) { 2569 error = ret; 2570 goto fail; 2571 } else 2572 cnx += ret; 2573 } 2574 2575 /* found states are sorted for outbound processing */ 2576 if (npols > 1) 2577 xfrm_state_sort(xfrm, tpp, cnx, family); 2578 2579 return cnx; 2580 2581 fail: 2582 for (cnx--; cnx >= 0; cnx--) 2583 xfrm_state_put(tpp[cnx]); 2584 return error; 2585 2586 } 2587 2588 static dscp_t xfrm_get_dscp(const struct flowi *fl, int family) 2589 { 2590 if (family == AF_INET) 2591 return fl->u.ip4.flowi4_dscp; 2592 2593 return 0; 2594 } 2595 2596 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family) 2597 { 2598 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 2599 struct dst_ops *dst_ops; 2600 struct xfrm_dst *xdst; 2601 2602 if (!afinfo) 2603 return ERR_PTR(-EINVAL); 2604 2605 switch (family) { 2606 case AF_INET: 2607 dst_ops = &net->xfrm.xfrm4_dst_ops; 2608 break; 2609 #if IS_ENABLED(CONFIG_IPV6) 2610 case AF_INET6: 2611 dst_ops = &net->xfrm.xfrm6_dst_ops; 2612 break; 2613 #endif 2614 default: 2615 BUG(); 2616 } 2617 xdst = dst_alloc(dst_ops, NULL, DST_OBSOLETE_NONE, 0); 2618 2619 if (likely(xdst)) { 2620 memset_after(xdst, 0, u.dst); 2621 } else 2622 xdst = ERR_PTR(-ENOBUFS); 2623 2624 rcu_read_unlock(); 2625 2626 return xdst; 2627 } 2628 2629 static void xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst, 2630 int nfheader_len) 2631 { 2632 if (dst->ops->family == AF_INET6) { 2633 path->path_cookie = rt6_get_cookie(dst_rt6_info(dst)); 2634 path->u.rt6.rt6i_nfheader_len = nfheader_len; 2635 } 2636 } 2637 2638 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev, 2639 const struct flowi *fl) 2640 { 2641 const struct xfrm_policy_afinfo *afinfo = 2642 xfrm_policy_get_afinfo(xdst->u.dst.ops->family); 2643 int err; 2644 2645 if (!afinfo) 2646 return -EINVAL; 2647 2648 err = afinfo->fill_dst(xdst, dev, fl); 2649 2650 rcu_read_unlock(); 2651 2652 return err; 2653 } 2654 2655 2656 /* Allocate chain of dst_entry's, attach known xfrm's, calculate 2657 * all the metrics... Shortly, bundle a bundle. 2658 */ 2659 2660 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy, 2661 struct xfrm_state **xfrm, 2662 struct xfrm_dst **bundle, 2663 int nx, 2664 const struct flowi *fl, 2665 struct dst_entry *dst) 2666 { 2667 const struct xfrm_state_afinfo *afinfo; 2668 const struct xfrm_mode *inner_mode; 2669 struct net *net = xp_net(policy); 2670 unsigned long now = jiffies; 2671 struct net_device *dev; 2672 struct xfrm_dst *xdst_prev = NULL; 2673 struct xfrm_dst *xdst0 = NULL; 2674 int i = 0; 2675 int err; 2676 int header_len = 0; 2677 int nfheader_len = 0; 2678 int trailer_len = 0; 2679 int family = policy->selector.family; 2680 xfrm_address_t saddr, daddr; 2681 dscp_t dscp; 2682 2683 xfrm_flowi_addr_get(fl, &saddr, &daddr, family); 2684 2685 dscp = xfrm_get_dscp(fl, family); 2686 2687 dst_hold(dst); 2688 2689 for (; i < nx; i++) { 2690 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family); 2691 struct dst_entry *dst1 = &xdst->u.dst; 2692 2693 err = PTR_ERR(xdst); 2694 if (IS_ERR(xdst)) { 2695 dst_release(dst); 2696 goto put_states; 2697 } 2698 2699 bundle[i] = xdst; 2700 if (!xdst_prev) 2701 xdst0 = xdst; 2702 else 2703 /* Ref count is taken during xfrm_alloc_dst() 2704 * No need to do dst_clone() on dst1 2705 */ 2706 xfrm_dst_set_child(xdst_prev, &xdst->u.dst); 2707 2708 if (xfrm[i]->sel.family == AF_UNSPEC) { 2709 inner_mode = xfrm_ip2inner_mode(xfrm[i], 2710 xfrm_af2proto(family)); 2711 if (!inner_mode) { 2712 err = -EAFNOSUPPORT; 2713 dst_release(dst); 2714 goto put_states; 2715 } 2716 } else 2717 inner_mode = &xfrm[i]->inner_mode; 2718 2719 xdst->route = dst; 2720 dst_copy_metrics(dst1, dst); 2721 2722 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) { 2723 __u32 mark = 0; 2724 int oif; 2725 2726 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m) 2727 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]); 2728 2729 if (xfrm[i]->xso.type != XFRM_DEV_OFFLOAD_PACKET) 2730 family = xfrm[i]->props.family; 2731 2732 oif = fl->flowi_oif ? : fl->flowi_l3mdev; 2733 dst = xfrm_dst_lookup(xfrm[i], dscp, oif, &saddr, 2734 &daddr, family, mark); 2735 err = PTR_ERR(dst); 2736 if (IS_ERR(dst)) 2737 goto put_states; 2738 } else 2739 dst_hold(dst); 2740 2741 dst1->xfrm = xfrm[i]; 2742 xdst->xfrm_genid = xfrm[i]->genid; 2743 2744 dst1->obsolete = DST_OBSOLETE_FORCE_CHK; 2745 dst1->lastuse = now; 2746 2747 dst1->input = dst_discard; 2748 2749 if (xfrm[i]->mode_cbs && xfrm[i]->mode_cbs->output) { 2750 dst1->output = xfrm[i]->mode_cbs->output; 2751 } else { 2752 rcu_read_lock(); 2753 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family); 2754 if (likely(afinfo)) 2755 dst1->output = afinfo->output; 2756 else 2757 dst1->output = dst_discard_out; 2758 rcu_read_unlock(); 2759 } 2760 2761 xdst_prev = xdst; 2762 2763 header_len += xfrm[i]->props.header_len; 2764 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT) 2765 nfheader_len += xfrm[i]->props.header_len; 2766 trailer_len += xfrm[i]->props.trailer_len; 2767 } 2768 2769 xfrm_dst_set_child(xdst_prev, dst); 2770 xdst0->path = dst; 2771 2772 err = -ENODEV; 2773 dev = dst->dev; 2774 if (!dev) 2775 goto free_dst; 2776 2777 xfrm_init_path(xdst0, dst, nfheader_len); 2778 xfrm_init_pmtu(bundle, nx); 2779 2780 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst; 2781 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) { 2782 err = xfrm_fill_dst(xdst_prev, dev, fl); 2783 if (err) 2784 goto free_dst; 2785 2786 xdst_prev->u.dst.header_len = header_len; 2787 xdst_prev->u.dst.trailer_len = trailer_len; 2788 header_len -= xdst_prev->u.dst.xfrm->props.header_len; 2789 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len; 2790 } 2791 2792 return &xdst0->u.dst; 2793 2794 put_states: 2795 for (; i < nx; i++) 2796 xfrm_state_put(xfrm[i]); 2797 free_dst: 2798 if (xdst0) 2799 dst_release_immediate(&xdst0->u.dst); 2800 2801 return ERR_PTR(err); 2802 } 2803 2804 static int xfrm_expand_policies(const struct flowi *fl, u16 family, 2805 struct xfrm_policy **pols, 2806 int *num_pols, int *num_xfrms) 2807 { 2808 int i; 2809 2810 if (*num_pols == 0 || !pols[0]) { 2811 *num_pols = 0; 2812 *num_xfrms = 0; 2813 return 0; 2814 } 2815 if (IS_ERR(pols[0])) { 2816 *num_pols = 0; 2817 return PTR_ERR(pols[0]); 2818 } 2819 2820 *num_xfrms = pols[0]->xfrm_nr; 2821 2822 #ifdef CONFIG_XFRM_SUB_POLICY 2823 if (pols[0]->action == XFRM_POLICY_ALLOW && 2824 pols[0]->type != XFRM_POLICY_TYPE_MAIN) { 2825 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]), 2826 XFRM_POLICY_TYPE_MAIN, 2827 fl, family, 2828 XFRM_POLICY_OUT, 2829 pols[0]->if_id); 2830 if (pols[1]) { 2831 if (IS_ERR(pols[1])) { 2832 xfrm_pols_put(pols, *num_pols); 2833 *num_pols = 0; 2834 return PTR_ERR(pols[1]); 2835 } 2836 (*num_pols)++; 2837 (*num_xfrms) += pols[1]->xfrm_nr; 2838 } 2839 } 2840 #endif 2841 for (i = 0; i < *num_pols; i++) { 2842 if (pols[i]->action != XFRM_POLICY_ALLOW) { 2843 *num_xfrms = -1; 2844 break; 2845 } 2846 } 2847 2848 return 0; 2849 2850 } 2851 2852 static struct xfrm_dst * 2853 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols, 2854 const struct flowi *fl, u16 family, 2855 struct dst_entry *dst_orig) 2856 { 2857 struct net *net = xp_net(pols[0]); 2858 struct xfrm_state *xfrm[XFRM_MAX_DEPTH]; 2859 struct xfrm_dst *bundle[XFRM_MAX_DEPTH]; 2860 struct xfrm_dst *xdst; 2861 struct dst_entry *dst; 2862 int err; 2863 2864 /* Try to instantiate a bundle */ 2865 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family); 2866 if (err <= 0) { 2867 if (err == 0) 2868 return NULL; 2869 2870 if (err != -EAGAIN) 2871 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR); 2872 return ERR_PTR(err); 2873 } 2874 2875 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig); 2876 if (IS_ERR(dst)) { 2877 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR); 2878 return ERR_CAST(dst); 2879 } 2880 2881 xdst = (struct xfrm_dst *)dst; 2882 xdst->num_xfrms = err; 2883 xdst->num_pols = num_pols; 2884 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols); 2885 xdst->policy_genid = atomic_read(&pols[0]->genid); 2886 2887 return xdst; 2888 } 2889 2890 static void xfrm_policy_queue_process(struct timer_list *t) 2891 { 2892 struct sk_buff *skb; 2893 struct sock *sk; 2894 struct dst_entry *dst; 2895 struct xfrm_policy *pol = timer_container_of(pol, t, polq.hold_timer); 2896 struct net *net = xp_net(pol); 2897 struct xfrm_policy_queue *pq = &pol->polq; 2898 struct flowi fl; 2899 struct sk_buff_head list; 2900 __u32 skb_mark; 2901 2902 spin_lock(&pq->hold_queue.lock); 2903 skb = skb_peek(&pq->hold_queue); 2904 if (!skb) { 2905 spin_unlock(&pq->hold_queue.lock); 2906 goto out; 2907 } 2908 dst = skb_dst(skb); 2909 sk = skb->sk; 2910 2911 /* Fixup the mark to support VTI. */ 2912 skb_mark = skb->mark; 2913 skb->mark = pol->mark.v; 2914 xfrm_decode_session(net, skb, &fl, dst->ops->family); 2915 skb->mark = skb_mark; 2916 spin_unlock(&pq->hold_queue.lock); 2917 2918 dst_hold(xfrm_dst_path(dst)); 2919 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE); 2920 if (IS_ERR(dst)) 2921 goto purge_queue; 2922 2923 if (dst->flags & DST_XFRM_QUEUE) { 2924 dst_release(dst); 2925 2926 if (pq->timeout >= XFRM_QUEUE_TMO_MAX) 2927 goto purge_queue; 2928 2929 pq->timeout = pq->timeout << 1; 2930 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout)) 2931 xfrm_pol_hold(pol); 2932 goto out; 2933 } 2934 2935 dst_release(dst); 2936 2937 __skb_queue_head_init(&list); 2938 2939 spin_lock(&pq->hold_queue.lock); 2940 pq->timeout = 0; 2941 skb_queue_splice_init(&pq->hold_queue, &list); 2942 spin_unlock(&pq->hold_queue.lock); 2943 2944 while (!skb_queue_empty(&list)) { 2945 skb = __skb_dequeue(&list); 2946 2947 /* Fixup the mark to support VTI. */ 2948 skb_mark = skb->mark; 2949 skb->mark = pol->mark.v; 2950 xfrm_decode_session(net, skb, &fl, skb_dst(skb)->ops->family); 2951 skb->mark = skb_mark; 2952 2953 dst_hold(xfrm_dst_path(skb_dst(skb))); 2954 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0); 2955 if (IS_ERR(dst)) { 2956 kfree_skb(skb); 2957 continue; 2958 } 2959 2960 nf_reset_ct(skb); 2961 skb_dst_drop(skb); 2962 skb_dst_set(skb, dst); 2963 2964 dst_output(net, skb_to_full_sk(skb), skb); 2965 } 2966 2967 out: 2968 xfrm_pol_put(pol); 2969 return; 2970 2971 purge_queue: 2972 pq->timeout = 0; 2973 skb_queue_purge(&pq->hold_queue); 2974 xfrm_pol_put(pol); 2975 } 2976 2977 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb) 2978 { 2979 unsigned long sched_next; 2980 struct dst_entry *dst = skb_dst(skb); 2981 struct xfrm_dst *xdst = (struct xfrm_dst *) dst; 2982 struct xfrm_policy *pol = xdst->pols[0]; 2983 struct xfrm_policy_queue *pq = &pol->polq; 2984 2985 if (unlikely(skb_fclone_busy(sk, skb))) { 2986 kfree_skb(skb); 2987 return 0; 2988 } 2989 2990 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) { 2991 kfree_skb(skb); 2992 return -EAGAIN; 2993 } 2994 2995 skb_dst_force(skb); 2996 2997 spin_lock_bh(&pq->hold_queue.lock); 2998 2999 if (!pq->timeout) 3000 pq->timeout = XFRM_QUEUE_TMO_MIN; 3001 3002 sched_next = jiffies + pq->timeout; 3003 3004 if (timer_delete(&pq->hold_timer)) { 3005 if (time_before(pq->hold_timer.expires, sched_next)) 3006 sched_next = pq->hold_timer.expires; 3007 xfrm_pol_put(pol); 3008 } 3009 3010 __skb_queue_tail(&pq->hold_queue, skb); 3011 if (!mod_timer(&pq->hold_timer, sched_next)) 3012 xfrm_pol_hold(pol); 3013 3014 spin_unlock_bh(&pq->hold_queue.lock); 3015 3016 return 0; 3017 } 3018 3019 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net, 3020 struct xfrm_flo *xflo, 3021 const struct flowi *fl, 3022 int num_xfrms, 3023 u16 family) 3024 { 3025 int err; 3026 struct net_device *dev; 3027 struct dst_entry *dst; 3028 struct dst_entry *dst1; 3029 struct xfrm_dst *xdst; 3030 3031 xdst = xfrm_alloc_dst(net, family); 3032 if (IS_ERR(xdst)) 3033 return xdst; 3034 3035 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) || 3036 net->xfrm.sysctl_larval_drop || 3037 num_xfrms <= 0) 3038 return xdst; 3039 3040 dst = xflo->dst_orig; 3041 dst1 = &xdst->u.dst; 3042 dst_hold(dst); 3043 xdst->route = dst; 3044 3045 dst_copy_metrics(dst1, dst); 3046 3047 dst1->obsolete = DST_OBSOLETE_FORCE_CHK; 3048 dst1->flags |= DST_XFRM_QUEUE; 3049 dst1->lastuse = jiffies; 3050 3051 dst1->input = dst_discard; 3052 dst1->output = xdst_queue_output; 3053 3054 dst_hold(dst); 3055 xfrm_dst_set_child(xdst, dst); 3056 xdst->path = dst; 3057 3058 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0); 3059 3060 err = -ENODEV; 3061 dev = dst->dev; 3062 if (!dev) 3063 goto free_dst; 3064 3065 err = xfrm_fill_dst(xdst, dev, fl); 3066 if (err) 3067 goto free_dst; 3068 3069 out: 3070 return xdst; 3071 3072 free_dst: 3073 dst_release(dst1); 3074 xdst = ERR_PTR(err); 3075 goto out; 3076 } 3077 3078 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net, 3079 const struct flowi *fl, 3080 u16 family, u8 dir, 3081 struct xfrm_flo *xflo, u32 if_id) 3082 { 3083 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 3084 int num_pols = 0, num_xfrms = 0, err; 3085 struct xfrm_dst *xdst; 3086 3087 /* Resolve policies to use if we couldn't get them from 3088 * previous cache entry */ 3089 num_pols = 1; 3090 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id); 3091 err = xfrm_expand_policies(fl, family, pols, 3092 &num_pols, &num_xfrms); 3093 if (err < 0) 3094 goto inc_error; 3095 if (num_pols == 0) 3096 return NULL; 3097 if (num_xfrms <= 0) 3098 goto make_dummy_bundle; 3099 3100 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, 3101 xflo->dst_orig); 3102 if (IS_ERR(xdst)) { 3103 err = PTR_ERR(xdst); 3104 if (err == -EREMOTE) { 3105 xfrm_pols_put(pols, num_pols); 3106 return NULL; 3107 } 3108 3109 if (err != -EAGAIN) 3110 goto error; 3111 goto make_dummy_bundle; 3112 } else if (xdst == NULL) { 3113 num_xfrms = 0; 3114 goto make_dummy_bundle; 3115 } 3116 3117 return xdst; 3118 3119 make_dummy_bundle: 3120 /* We found policies, but there's no bundles to instantiate: 3121 * either because the policy blocks, has no transformations or 3122 * we could not build template (no xfrm_states).*/ 3123 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family); 3124 if (IS_ERR(xdst)) { 3125 xfrm_pols_put(pols, num_pols); 3126 return ERR_CAST(xdst); 3127 } 3128 xdst->num_pols = num_pols; 3129 xdst->num_xfrms = num_xfrms; 3130 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols); 3131 3132 return xdst; 3133 3134 inc_error: 3135 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR); 3136 error: 3137 xfrm_pols_put(pols, num_pols); 3138 return ERR_PTR(err); 3139 } 3140 3141 static struct dst_entry *make_blackhole(struct net *net, u16 family, 3142 struct dst_entry *dst_orig) 3143 { 3144 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 3145 struct dst_entry *ret; 3146 3147 if (!afinfo) { 3148 dst_release(dst_orig); 3149 return ERR_PTR(-EINVAL); 3150 } else { 3151 ret = afinfo->blackhole_route(net, dst_orig); 3152 } 3153 rcu_read_unlock(); 3154 3155 return ret; 3156 } 3157 3158 /* Finds/creates a bundle for given flow and if_id 3159 * 3160 * At the moment we eat a raw IP route. Mostly to speed up lookups 3161 * on interfaces with disabled IPsec. 3162 * 3163 * xfrm_lookup uses an if_id of 0 by default, and is provided for 3164 * compatibility 3165 */ 3166 struct dst_entry *xfrm_lookup_with_ifid(struct net *net, 3167 struct dst_entry *dst_orig, 3168 const struct flowi *fl, 3169 const struct sock *sk, 3170 int flags, u32 if_id) 3171 { 3172 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 3173 struct xfrm_dst *xdst; 3174 struct dst_entry *dst, *route; 3175 u16 family = dst_orig->ops->family; 3176 u8 dir = XFRM_POLICY_OUT; 3177 int i, err, num_pols, num_xfrms = 0, drop_pols = 0; 3178 3179 dst = NULL; 3180 xdst = NULL; 3181 route = NULL; 3182 3183 sk = sk_const_to_full_sk(sk); 3184 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) { 3185 num_pols = 1; 3186 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family, 3187 if_id); 3188 err = xfrm_expand_policies(fl, family, pols, 3189 &num_pols, &num_xfrms); 3190 if (err < 0) 3191 goto dropdst; 3192 3193 if (num_pols) { 3194 if (num_xfrms <= 0) { 3195 drop_pols = num_pols; 3196 goto no_transform; 3197 } 3198 3199 xdst = xfrm_resolve_and_create_bundle( 3200 pols, num_pols, fl, 3201 family, dst_orig); 3202 3203 if (IS_ERR(xdst)) { 3204 xfrm_pols_put(pols, num_pols); 3205 err = PTR_ERR(xdst); 3206 if (err == -EREMOTE) 3207 goto nopol; 3208 3209 goto dropdst; 3210 } else if (xdst == NULL) { 3211 num_xfrms = 0; 3212 drop_pols = num_pols; 3213 goto no_transform; 3214 } 3215 3216 route = xdst->route; 3217 } 3218 } 3219 3220 if (xdst == NULL) { 3221 struct xfrm_flo xflo; 3222 3223 xflo.dst_orig = dst_orig; 3224 xflo.flags = flags; 3225 3226 /* To accelerate a bit... */ 3227 if (!if_id && ((dst_orig->flags & DST_NOXFRM) || 3228 !READ_ONCE(net->xfrm.policy_count[XFRM_POLICY_OUT]))) 3229 goto nopol; 3230 3231 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id); 3232 if (xdst == NULL) 3233 goto nopol; 3234 if (IS_ERR(xdst)) { 3235 err = PTR_ERR(xdst); 3236 goto dropdst; 3237 } 3238 3239 num_pols = xdst->num_pols; 3240 num_xfrms = xdst->num_xfrms; 3241 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols); 3242 route = xdst->route; 3243 } 3244 3245 dst = &xdst->u.dst; 3246 if (route == NULL && num_xfrms > 0) { 3247 /* The only case when xfrm_bundle_lookup() returns a 3248 * bundle with null route, is when the template could 3249 * not be resolved. It means policies are there, but 3250 * bundle could not be created, since we don't yet 3251 * have the xfrm_state's. We need to wait for KM to 3252 * negotiate new SA's or bail out with error.*/ 3253 if (net->xfrm.sysctl_larval_drop) { 3254 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES); 3255 err = -EREMOTE; 3256 goto error; 3257 } 3258 3259 err = -EAGAIN; 3260 3261 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES); 3262 goto error; 3263 } 3264 3265 no_transform: 3266 if (num_pols == 0) 3267 goto nopol; 3268 3269 if ((flags & XFRM_LOOKUP_ICMP) && 3270 !(pols[0]->flags & XFRM_POLICY_ICMP)) { 3271 err = -ENOENT; 3272 goto error; 3273 } 3274 3275 for (i = 0; i < num_pols; i++) 3276 WRITE_ONCE(pols[i]->curlft.use_time, ktime_get_real_seconds()); 3277 3278 if (num_xfrms < 0) { 3279 /* Prohibit the flow */ 3280 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK); 3281 err = -EPERM; 3282 goto error; 3283 } else if (num_xfrms > 0) { 3284 /* Flow transformed */ 3285 dst_release(dst_orig); 3286 } else { 3287 /* Flow passes untransformed */ 3288 dst_release(dst); 3289 dst = dst_orig; 3290 } 3291 3292 ok: 3293 xfrm_pols_put(pols, drop_pols); 3294 if (dst->xfrm && 3295 (dst->xfrm->props.mode == XFRM_MODE_TUNNEL || 3296 dst->xfrm->props.mode == XFRM_MODE_IPTFS)) 3297 dst->flags |= DST_XFRM_TUNNEL; 3298 return dst; 3299 3300 nopol: 3301 if ((!dst_orig->dev || !(dst_orig->dev->flags & IFF_LOOPBACK)) && 3302 READ_ONCE(net->xfrm.policy_default[dir]) == XFRM_USERPOLICY_BLOCK) { 3303 err = -EPERM; 3304 goto error; 3305 } 3306 if (!(flags & XFRM_LOOKUP_ICMP)) { 3307 dst = dst_orig; 3308 goto ok; 3309 } 3310 err = -ENOENT; 3311 error: 3312 dst_release(dst); 3313 dropdst: 3314 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF)) 3315 dst_release(dst_orig); 3316 xfrm_pols_put(pols, drop_pols); 3317 return ERR_PTR(err); 3318 } 3319 EXPORT_SYMBOL(xfrm_lookup_with_ifid); 3320 3321 /* Main function: finds/creates a bundle for given flow. 3322 * 3323 * At the moment we eat a raw IP route. Mostly to speed up lookups 3324 * on interfaces with disabled IPsec. 3325 */ 3326 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig, 3327 const struct flowi *fl, const struct sock *sk, 3328 int flags) 3329 { 3330 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0); 3331 } 3332 EXPORT_SYMBOL(xfrm_lookup); 3333 3334 /* Callers of xfrm_lookup_route() must ensure a call to dst_output(). 3335 * Otherwise we may send out blackholed packets. 3336 */ 3337 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig, 3338 const struct flowi *fl, 3339 const struct sock *sk, int flags) 3340 { 3341 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk, 3342 flags | XFRM_LOOKUP_QUEUE | 3343 XFRM_LOOKUP_KEEP_DST_REF); 3344 3345 if (PTR_ERR(dst) == -EREMOTE) 3346 return make_blackhole(net, dst_orig->ops->family, dst_orig); 3347 3348 if (IS_ERR(dst)) 3349 dst_release(dst_orig); 3350 3351 return dst; 3352 } 3353 EXPORT_SYMBOL(xfrm_lookup_route); 3354 3355 static inline int 3356 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl) 3357 { 3358 struct sec_path *sp = skb_sec_path(skb); 3359 struct xfrm_state *x; 3360 3361 if (!sp || idx < 0 || idx >= sp->len) 3362 return 0; 3363 x = sp->xvec[idx]; 3364 if (!x->type->reject) 3365 return 0; 3366 return x->type->reject(x, skb, fl); 3367 } 3368 3369 /* When skb is transformed back to its "native" form, we have to 3370 * check policy restrictions. At the moment we make this in maximally 3371 * stupid way. Shame on me. :-) Of course, connected sockets must 3372 * have policy cached at them. 3373 */ 3374 3375 static inline int 3376 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x, 3377 unsigned short family, u32 if_id) 3378 { 3379 if (xfrm_state_kern(x)) 3380 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family); 3381 return x->id.proto == tmpl->id.proto && 3382 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) && 3383 (x->props.reqid == tmpl->reqid || !tmpl->reqid) && 3384 x->props.mode == tmpl->mode && 3385 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) || 3386 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) && 3387 !(x->props.mode != XFRM_MODE_TRANSPORT && 3388 xfrm_state_addr_cmp(tmpl, x, family)) && 3389 (if_id == 0 || if_id == x->if_id); 3390 } 3391 3392 /* 3393 * 0 or more than 0 is returned when validation is succeeded (either bypass 3394 * because of optional transport mode, or next index of the matched secpath 3395 * state with the template. 3396 * -1 is returned when no matching template is found. 3397 * Otherwise "-2 - errored_index" is returned. 3398 */ 3399 static inline int 3400 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start, 3401 unsigned short family, u32 if_id) 3402 { 3403 int idx = start; 3404 3405 if (tmpl->optional) { 3406 if (tmpl->mode == XFRM_MODE_TRANSPORT) 3407 return start; 3408 } else 3409 start = -1; 3410 for (; idx < sp->len; idx++) { 3411 if (xfrm_state_ok(tmpl, sp->xvec[idx], family, if_id)) 3412 return ++idx; 3413 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) { 3414 if (idx < sp->verified_cnt) { 3415 /* Secpath entry previously verified, consider optional and 3416 * continue searching 3417 */ 3418 continue; 3419 } 3420 3421 if (start == -1) 3422 start = -2-idx; 3423 break; 3424 } 3425 } 3426 return start; 3427 } 3428 3429 static void 3430 decode_session4(const struct xfrm_flow_keys *flkeys, struct flowi *fl, bool reverse) 3431 { 3432 struct flowi4 *fl4 = &fl->u.ip4; 3433 3434 memset(fl4, 0, sizeof(struct flowi4)); 3435 3436 if (reverse) { 3437 fl4->saddr = flkeys->addrs.ipv4.dst; 3438 fl4->daddr = flkeys->addrs.ipv4.src; 3439 fl4->fl4_sport = flkeys->ports.dst; 3440 fl4->fl4_dport = flkeys->ports.src; 3441 } else { 3442 fl4->saddr = flkeys->addrs.ipv4.src; 3443 fl4->daddr = flkeys->addrs.ipv4.dst; 3444 fl4->fl4_sport = flkeys->ports.src; 3445 fl4->fl4_dport = flkeys->ports.dst; 3446 } 3447 3448 switch (flkeys->basic.ip_proto) { 3449 case IPPROTO_GRE: 3450 fl4->fl4_gre_key = flkeys->gre.keyid; 3451 break; 3452 case IPPROTO_ICMP: 3453 fl4->fl4_icmp_type = flkeys->icmp.type; 3454 fl4->fl4_icmp_code = flkeys->icmp.code; 3455 break; 3456 } 3457 3458 fl4->flowi4_proto = flkeys->basic.ip_proto; 3459 fl4->flowi4_dscp = inet_dsfield_to_dscp(flkeys->ip.tos); 3460 } 3461 3462 #if IS_ENABLED(CONFIG_IPV6) 3463 static void 3464 decode_session6(const struct xfrm_flow_keys *flkeys, struct flowi *fl, bool reverse) 3465 { 3466 struct flowi6 *fl6 = &fl->u.ip6; 3467 3468 memset(fl6, 0, sizeof(struct flowi6)); 3469 3470 if (reverse) { 3471 fl6->saddr = flkeys->addrs.ipv6.dst; 3472 fl6->daddr = flkeys->addrs.ipv6.src; 3473 fl6->fl6_sport = flkeys->ports.dst; 3474 fl6->fl6_dport = flkeys->ports.src; 3475 } else { 3476 fl6->saddr = flkeys->addrs.ipv6.src; 3477 fl6->daddr = flkeys->addrs.ipv6.dst; 3478 fl6->fl6_sport = flkeys->ports.src; 3479 fl6->fl6_dport = flkeys->ports.dst; 3480 } 3481 3482 switch (flkeys->basic.ip_proto) { 3483 case IPPROTO_GRE: 3484 fl6->fl6_gre_key = flkeys->gre.keyid; 3485 break; 3486 case IPPROTO_ICMPV6: 3487 fl6->fl6_icmp_type = flkeys->icmp.type; 3488 fl6->fl6_icmp_code = flkeys->icmp.code; 3489 break; 3490 } 3491 3492 fl6->flowi6_proto = flkeys->basic.ip_proto; 3493 } 3494 #endif 3495 3496 int __xfrm_decode_session(struct net *net, struct sk_buff *skb, struct flowi *fl, 3497 unsigned int family, int reverse) 3498 { 3499 struct xfrm_flow_keys flkeys; 3500 3501 memset(&flkeys, 0, sizeof(flkeys)); 3502 __skb_flow_dissect(net, skb, &xfrm_session_dissector, &flkeys, 3503 NULL, 0, 0, 0, FLOW_DISSECTOR_F_STOP_AT_ENCAP); 3504 3505 switch (family) { 3506 case AF_INET: 3507 decode_session4(&flkeys, fl, reverse); 3508 break; 3509 #if IS_ENABLED(CONFIG_IPV6) 3510 case AF_INET6: 3511 decode_session6(&flkeys, fl, reverse); 3512 break; 3513 #endif 3514 default: 3515 return -EAFNOSUPPORT; 3516 } 3517 3518 fl->flowi_mark = skb->mark; 3519 if (reverse) { 3520 fl->flowi_oif = skb->skb_iif; 3521 } else { 3522 int oif = 0; 3523 3524 if (skb_dst(skb) && skb_dst(skb)->dev) 3525 oif = skb_dst(skb)->dev->ifindex; 3526 3527 fl->flowi_oif = oif; 3528 } 3529 3530 return security_xfrm_decode_session(skb, &fl->flowi_secid); 3531 } 3532 EXPORT_SYMBOL(__xfrm_decode_session); 3533 3534 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp) 3535 { 3536 for (; k < sp->len; k++) { 3537 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) { 3538 *idxp = k; 3539 return 1; 3540 } 3541 } 3542 3543 return 0; 3544 } 3545 3546 static bool icmp_err_packet(const struct flowi *fl, unsigned short family) 3547 { 3548 const struct flowi4 *fl4 = &fl->u.ip4; 3549 3550 if (family == AF_INET && 3551 fl4->flowi4_proto == IPPROTO_ICMP && 3552 (fl4->fl4_icmp_type == ICMP_DEST_UNREACH || 3553 fl4->fl4_icmp_type == ICMP_TIME_EXCEEDED)) 3554 return true; 3555 3556 #if IS_ENABLED(CONFIG_IPV6) 3557 if (family == AF_INET6) { 3558 const struct flowi6 *fl6 = &fl->u.ip6; 3559 3560 if (fl6->flowi6_proto == IPPROTO_ICMPV6 && 3561 (fl6->fl6_icmp_type == ICMPV6_DEST_UNREACH || 3562 fl6->fl6_icmp_type == ICMPV6_PKT_TOOBIG || 3563 fl6->fl6_icmp_type == ICMPV6_TIME_EXCEED)) 3564 return true; 3565 } 3566 #endif 3567 return false; 3568 } 3569 3570 static bool xfrm_icmp_flow_decode(struct sk_buff *skb, unsigned short family, 3571 const struct flowi *fl, struct flowi *fl1) 3572 { 3573 bool ret = true; 3574 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 3575 int hl = family == AF_INET ? (sizeof(struct iphdr) + sizeof(struct icmphdr)) : 3576 (sizeof(struct ipv6hdr) + sizeof(struct icmp6hdr)); 3577 3578 if (!newskb) 3579 return true; 3580 3581 if (!pskb_pull(newskb, hl)) 3582 goto out; 3583 3584 skb_reset_network_header(newskb); 3585 3586 if (xfrm_decode_session_reverse(dev_net(skb->dev), newskb, fl1, family) < 0) 3587 goto out; 3588 3589 fl1->flowi_oif = fl->flowi_oif; 3590 fl1->flowi_mark = fl->flowi_mark; 3591 fl1->flowi_dscp = fl->flowi_dscp; 3592 nf_nat_decode_session(newskb, fl1, family); 3593 ret = false; 3594 3595 out: 3596 consume_skb(newskb); 3597 return ret; 3598 } 3599 3600 static bool xfrm_selector_inner_icmp_match(struct sk_buff *skb, unsigned short family, 3601 const struct xfrm_selector *sel, 3602 const struct flowi *fl) 3603 { 3604 bool ret = false; 3605 3606 if (icmp_err_packet(fl, family)) { 3607 struct flowi fl1; 3608 3609 if (xfrm_icmp_flow_decode(skb, family, fl, &fl1)) 3610 return ret; 3611 3612 ret = xfrm_selector_match(sel, &fl1, family); 3613 } 3614 3615 return ret; 3616 } 3617 3618 static inline struct 3619 xfrm_policy *xfrm_in_fwd_icmp(struct sk_buff *skb, 3620 const struct flowi *fl, unsigned short family, 3621 u32 if_id) 3622 { 3623 struct xfrm_policy *pol = NULL; 3624 3625 if (icmp_err_packet(fl, family)) { 3626 struct flowi fl1; 3627 struct net *net = dev_net(skb->dev); 3628 3629 if (xfrm_icmp_flow_decode(skb, family, fl, &fl1)) 3630 return pol; 3631 3632 pol = xfrm_policy_lookup(net, &fl1, family, XFRM_POLICY_FWD, if_id); 3633 if (IS_ERR(pol)) 3634 pol = NULL; 3635 } 3636 3637 return pol; 3638 } 3639 3640 static inline struct 3641 dst_entry *xfrm_out_fwd_icmp(struct sk_buff *skb, struct flowi *fl, 3642 unsigned short family, struct dst_entry *dst) 3643 { 3644 if (icmp_err_packet(fl, family)) { 3645 struct net *net = dev_net(skb->dev); 3646 struct dst_entry *dst2; 3647 struct flowi fl1; 3648 3649 if (xfrm_icmp_flow_decode(skb, family, fl, &fl1)) 3650 return dst; 3651 3652 dst_hold(dst); 3653 3654 dst2 = xfrm_lookup(net, dst, &fl1, NULL, (XFRM_LOOKUP_QUEUE | XFRM_LOOKUP_ICMP)); 3655 3656 if (IS_ERR(dst2)) 3657 return dst; 3658 3659 if (dst2->xfrm) { 3660 dst_release(dst); 3661 dst = dst2; 3662 } else { 3663 dst_release(dst2); 3664 } 3665 } 3666 3667 return dst; 3668 } 3669 3670 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 3671 unsigned short family) 3672 { 3673 struct net *net = dev_net(skb->dev); 3674 struct xfrm_policy *pol; 3675 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 3676 int npols = 0; 3677 int xfrm_nr; 3678 int pi; 3679 int reverse; 3680 struct flowi fl; 3681 int xerr_idx = -1; 3682 const struct xfrm_if_cb *ifcb; 3683 struct sec_path *sp; 3684 u32 if_id = 0; 3685 3686 rcu_read_lock(); 3687 ifcb = xfrm_if_get_cb(); 3688 3689 if (ifcb) { 3690 struct xfrm_if_decode_session_result r; 3691 3692 if (ifcb->decode_session(skb, family, &r)) { 3693 if_id = r.if_id; 3694 net = r.net; 3695 } 3696 } 3697 rcu_read_unlock(); 3698 3699 reverse = dir & ~XFRM_POLICY_MASK; 3700 dir &= XFRM_POLICY_MASK; 3701 3702 if (__xfrm_decode_session(net, skb, &fl, family, reverse) < 0) { 3703 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR); 3704 return 0; 3705 } 3706 3707 nf_nat_decode_session(skb, &fl, family); 3708 3709 /* First, check used SA against their selectors. */ 3710 sp = skb_sec_path(skb); 3711 if (sp) { 3712 int i; 3713 3714 for (i = sp->len - 1; i >= 0; i--) { 3715 struct xfrm_state *x = sp->xvec[i]; 3716 int ret = 0; 3717 3718 if (!xfrm_selector_match(&x->sel, &fl, family)) { 3719 ret = 1; 3720 if (x->props.flags & XFRM_STATE_ICMP && 3721 xfrm_selector_inner_icmp_match(skb, family, &x->sel, &fl)) 3722 ret = 0; 3723 if (ret) { 3724 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH); 3725 return 0; 3726 } 3727 } 3728 } 3729 } 3730 3731 pol = NULL; 3732 sk = sk_to_full_sk(sk); 3733 if (sk && sk->sk_policy[dir]) { 3734 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id); 3735 if (IS_ERR(pol)) { 3736 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3737 return 0; 3738 } 3739 } 3740 3741 if (!pol) 3742 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id); 3743 3744 if (IS_ERR(pol)) { 3745 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3746 return 0; 3747 } 3748 3749 if (!pol && dir == XFRM_POLICY_FWD) 3750 pol = xfrm_in_fwd_icmp(skb, &fl, family, if_id); 3751 3752 if (!pol) { 3753 const bool is_crypto_offload = sp && 3754 (xfrm_input_state(skb)->xso.type == XFRM_DEV_OFFLOAD_CRYPTO); 3755 3756 if (READ_ONCE(net->xfrm.policy_default[dir]) == XFRM_USERPOLICY_BLOCK) { 3757 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS); 3758 return 0; 3759 } 3760 3761 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx) && !is_crypto_offload) { 3762 xfrm_secpath_reject(xerr_idx, skb, &fl); 3763 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS); 3764 return 0; 3765 } 3766 return 1; 3767 } 3768 3769 /* This lockless write can happen from different cpus. */ 3770 WRITE_ONCE(pol->curlft.use_time, ktime_get_real_seconds()); 3771 3772 pols[0] = pol; 3773 npols++; 3774 #ifdef CONFIG_XFRM_SUB_POLICY 3775 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) { 3776 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, 3777 &fl, family, 3778 XFRM_POLICY_IN, if_id); 3779 if (pols[1]) { 3780 if (IS_ERR(pols[1])) { 3781 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3782 xfrm_pol_put(pols[0]); 3783 return 0; 3784 } 3785 /* This write can happen from different cpus. */ 3786 WRITE_ONCE(pols[1]->curlft.use_time, 3787 ktime_get_real_seconds()); 3788 npols++; 3789 } 3790 } 3791 #endif 3792 3793 if (pol->action == XFRM_POLICY_ALLOW) { 3794 static struct sec_path dummy; 3795 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH]; 3796 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH]; 3797 struct xfrm_tmpl **tpp = tp; 3798 int i, k = 0; 3799 int ti = 0; 3800 3801 sp = skb_sec_path(skb); 3802 if (!sp) 3803 sp = &dummy; 3804 3805 for (pi = 0; pi < npols; pi++) { 3806 if (pols[pi] != pol && 3807 pols[pi]->action != XFRM_POLICY_ALLOW) { 3808 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK); 3809 goto reject; 3810 } 3811 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) { 3812 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR); 3813 goto reject_error; 3814 } 3815 for (i = 0; i < pols[pi]->xfrm_nr; i++) 3816 tpp[ti++] = &pols[pi]->xfrm_vec[i]; 3817 } 3818 xfrm_nr = ti; 3819 3820 if (npols > 1) { 3821 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family); 3822 tpp = stp; 3823 } 3824 3825 if (pol->xdo.type == XFRM_DEV_OFFLOAD_PACKET && sp == &dummy) 3826 /* This policy template was already checked by HW 3827 * and secpath was removed in __xfrm_policy_check2. 3828 */ 3829 goto out; 3830 3831 /* For each tunnel xfrm, find the first matching tmpl. 3832 * For each tmpl before that, find corresponding xfrm. 3833 * Order is _important_. Later we will implement 3834 * some barriers, but at the moment barriers 3835 * are implied between each two transformations. 3836 * Upon success, marks secpath entries as having been 3837 * verified to allow them to be skipped in future policy 3838 * checks (e.g. nested tunnels). 3839 */ 3840 for (i = xfrm_nr - 1; i >= 0; i--) { 3841 k = xfrm_policy_ok(tpp[i], sp, k, family, if_id); 3842 if (k < 0) { 3843 if (k < -1) 3844 /* "-2 - errored_index" returned */ 3845 xerr_idx = -(2+k); 3846 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH); 3847 goto reject; 3848 } 3849 } 3850 3851 if (secpath_has_nontransport(sp, k, &xerr_idx)) { 3852 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH); 3853 goto reject; 3854 } 3855 3856 out: 3857 xfrm_pols_put(pols, npols); 3858 sp->verified_cnt = k; 3859 3860 return 1; 3861 } 3862 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK); 3863 3864 reject: 3865 xfrm_secpath_reject(xerr_idx, skb, &fl); 3866 reject_error: 3867 xfrm_pols_put(pols, npols); 3868 return 0; 3869 } 3870 EXPORT_SYMBOL(__xfrm_policy_check); 3871 3872 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family) 3873 { 3874 struct net *net = dev_net(skb->dev); 3875 struct flowi fl; 3876 struct dst_entry *dst; 3877 int res = 1; 3878 3879 if (xfrm_decode_session(net, skb, &fl, family) < 0) { 3880 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR); 3881 return 0; 3882 } 3883 3884 skb_dst_force(skb); 3885 dst = skb_dst(skb); 3886 if (!dst) { 3887 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR); 3888 return 0; 3889 } 3890 3891 /* ignore return value from skb_dstref_steal, xfrm_lookup takes 3892 * care of dropping the refcnt if needed. 3893 */ 3894 skb_dstref_steal(skb); 3895 3896 dst = xfrm_lookup(net, dst, &fl, NULL, XFRM_LOOKUP_QUEUE); 3897 if (IS_ERR(dst)) { 3898 res = 0; 3899 dst = NULL; 3900 } 3901 3902 if (dst && !dst->xfrm) 3903 dst = xfrm_out_fwd_icmp(skb, &fl, family, dst); 3904 3905 skb_dst_set(skb, dst); 3906 return res; 3907 } 3908 EXPORT_SYMBOL(__xfrm_route_forward); 3909 3910 /* Optimize later using cookies and generation ids. */ 3911 3912 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie) 3913 { 3914 /* Code (such as xfrm_bundle_create()) sets dst->obsolete 3915 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to 3916 * get validated by dst_ops->check on every use. We do this 3917 * because when a normal route referenced by an XFRM dst is 3918 * obsoleted we do not go looking around for all parent 3919 * referencing XFRM dsts so that we can invalidate them. It 3920 * is just too much work. Instead we make the checks here on 3921 * every use. For example: 3922 * 3923 * XFRM dst A --> IPv4 dst X 3924 * 3925 * X is the "xdst->route" of A (X is also the "dst->path" of A 3926 * in this example). If X is marked obsolete, "A" will not 3927 * notice. That's what we are validating here via the 3928 * stale_bundle() check. 3929 * 3930 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will 3931 * be marked on it. 3932 * This will force stale_bundle() to fail on any xdst bundle with 3933 * this dst linked in it. 3934 */ 3935 if (READ_ONCE(dst->obsolete) < 0 && !stale_bundle(dst)) 3936 return dst; 3937 3938 return NULL; 3939 } 3940 3941 static int stale_bundle(struct dst_entry *dst) 3942 { 3943 return !xfrm_bundle_ok((struct xfrm_dst *)dst); 3944 } 3945 3946 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev) 3947 { 3948 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) { 3949 dst->dev = blackhole_netdev; 3950 dev_hold(dst->dev); 3951 dev_put(dev); 3952 } 3953 } 3954 EXPORT_SYMBOL(xfrm_dst_ifdown); 3955 3956 static void xfrm_link_failure(struct sk_buff *skb) 3957 { 3958 /* Impossible. Such dst must be popped before reaches point of failure. */ 3959 } 3960 3961 static void xfrm_negative_advice(struct sock *sk, struct dst_entry *dst) 3962 { 3963 if (READ_ONCE(dst->obsolete)) 3964 sk_dst_reset(sk); 3965 } 3966 3967 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr) 3968 { 3969 while (nr--) { 3970 struct xfrm_dst *xdst = bundle[nr]; 3971 u32 pmtu, route_mtu_cached; 3972 struct dst_entry *dst; 3973 3974 dst = &xdst->u.dst; 3975 pmtu = dst_mtu(xfrm_dst_child(dst)); 3976 xdst->child_mtu_cached = pmtu; 3977 3978 pmtu = xfrm_state_mtu(dst->xfrm, pmtu); 3979 3980 route_mtu_cached = dst_mtu(xdst->route); 3981 xdst->route_mtu_cached = route_mtu_cached; 3982 3983 if (pmtu > route_mtu_cached) 3984 pmtu = route_mtu_cached; 3985 3986 dst_metric_set(dst, RTAX_MTU, pmtu); 3987 } 3988 } 3989 3990 /* Check that the bundle accepts the flow and its components are 3991 * still valid. 3992 */ 3993 3994 static int xfrm_bundle_ok(struct xfrm_dst *first) 3995 { 3996 struct xfrm_dst *bundle[XFRM_MAX_DEPTH]; 3997 struct dst_entry *dst = &first->u.dst; 3998 struct xfrm_dst *xdst; 3999 int start_from, nr; 4000 u32 mtu; 4001 4002 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) || 4003 (dst->dev && !netif_running(dst->dev))) 4004 return 0; 4005 4006 if (dst->flags & DST_XFRM_QUEUE) 4007 return 1; 4008 4009 start_from = nr = 0; 4010 do { 4011 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 4012 4013 if (dst->xfrm->km.state != XFRM_STATE_VALID) 4014 return 0; 4015 if (xdst->xfrm_genid != dst->xfrm->genid) 4016 return 0; 4017 if (xdst->num_pols > 0 && 4018 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid)) 4019 return 0; 4020 4021 bundle[nr++] = xdst; 4022 4023 mtu = dst_mtu(xfrm_dst_child(dst)); 4024 if (xdst->child_mtu_cached != mtu) { 4025 start_from = nr; 4026 xdst->child_mtu_cached = mtu; 4027 } 4028 4029 if (!dst_check(xdst->route, xdst->route_cookie)) 4030 return 0; 4031 mtu = dst_mtu(xdst->route); 4032 if (xdst->route_mtu_cached != mtu) { 4033 start_from = nr; 4034 xdst->route_mtu_cached = mtu; 4035 } 4036 4037 dst = xfrm_dst_child(dst); 4038 } while (dst->xfrm); 4039 4040 if (likely(!start_from)) 4041 return 1; 4042 4043 xdst = bundle[start_from - 1]; 4044 mtu = xdst->child_mtu_cached; 4045 while (start_from--) { 4046 dst = &xdst->u.dst; 4047 4048 mtu = xfrm_state_mtu(dst->xfrm, mtu); 4049 if (mtu > xdst->route_mtu_cached) 4050 mtu = xdst->route_mtu_cached; 4051 dst_metric_set(dst, RTAX_MTU, mtu); 4052 if (!start_from) 4053 break; 4054 4055 xdst = bundle[start_from - 1]; 4056 xdst->child_mtu_cached = mtu; 4057 } 4058 4059 return 1; 4060 } 4061 4062 static unsigned int xfrm_default_advmss(const struct dst_entry *dst) 4063 { 4064 return dst_metric_advmss(xfrm_dst_path(dst)); 4065 } 4066 4067 static unsigned int xfrm_mtu(const struct dst_entry *dst) 4068 { 4069 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 4070 4071 return mtu ? : dst_mtu(xfrm_dst_path(dst)); 4072 } 4073 4074 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst, 4075 const void *daddr) 4076 { 4077 while (dst->xfrm) { 4078 const struct xfrm_state *xfrm = dst->xfrm; 4079 4080 dst = xfrm_dst_child(dst); 4081 4082 if (xfrm->props.mode == XFRM_MODE_TRANSPORT) 4083 continue; 4084 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR) 4085 daddr = xfrm->coaddr; 4086 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR)) 4087 daddr = &xfrm->id.daddr; 4088 } 4089 return daddr; 4090 } 4091 4092 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst, 4093 struct sk_buff *skb, 4094 const void *daddr) 4095 { 4096 const struct dst_entry *path = xfrm_dst_path(dst); 4097 4098 if (!skb) 4099 daddr = xfrm_get_dst_nexthop(dst, daddr); 4100 return path->ops->neigh_lookup(path, skb, daddr); 4101 } 4102 4103 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr) 4104 { 4105 const struct dst_entry *path = xfrm_dst_path(dst); 4106 4107 daddr = xfrm_get_dst_nexthop(dst, daddr); 4108 path->ops->confirm_neigh(path, daddr); 4109 } 4110 4111 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family) 4112 { 4113 int err = 0; 4114 4115 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo))) 4116 return -EAFNOSUPPORT; 4117 4118 spin_lock(&xfrm_policy_afinfo_lock); 4119 if (unlikely(xfrm_policy_afinfo[family] != NULL)) 4120 err = -EEXIST; 4121 else { 4122 struct dst_ops *dst_ops = afinfo->dst_ops; 4123 if (likely(dst_ops->kmem_cachep == NULL)) 4124 dst_ops->kmem_cachep = xfrm_dst_cache; 4125 if (likely(dst_ops->check == NULL)) 4126 dst_ops->check = xfrm_dst_check; 4127 if (likely(dst_ops->default_advmss == NULL)) 4128 dst_ops->default_advmss = xfrm_default_advmss; 4129 if (likely(dst_ops->mtu == NULL)) 4130 dst_ops->mtu = xfrm_mtu; 4131 if (likely(dst_ops->negative_advice == NULL)) 4132 dst_ops->negative_advice = xfrm_negative_advice; 4133 if (likely(dst_ops->link_failure == NULL)) 4134 dst_ops->link_failure = xfrm_link_failure; 4135 if (likely(dst_ops->neigh_lookup == NULL)) 4136 dst_ops->neigh_lookup = xfrm_neigh_lookup; 4137 if (likely(!dst_ops->confirm_neigh)) 4138 dst_ops->confirm_neigh = xfrm_confirm_neigh; 4139 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo); 4140 } 4141 spin_unlock(&xfrm_policy_afinfo_lock); 4142 4143 return err; 4144 } 4145 EXPORT_SYMBOL(xfrm_policy_register_afinfo); 4146 4147 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo) 4148 { 4149 struct dst_ops *dst_ops = afinfo->dst_ops; 4150 int i; 4151 4152 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) { 4153 if (rcu_access_pointer(xfrm_policy_afinfo[i]) != afinfo) 4154 continue; 4155 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL); 4156 break; 4157 } 4158 4159 synchronize_rcu(); 4160 4161 dst_ops->kmem_cachep = NULL; 4162 dst_ops->check = NULL; 4163 dst_ops->negative_advice = NULL; 4164 dst_ops->link_failure = NULL; 4165 } 4166 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo); 4167 4168 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb) 4169 { 4170 spin_lock(&xfrm_if_cb_lock); 4171 rcu_assign_pointer(xfrm_if_cb, ifcb); 4172 spin_unlock(&xfrm_if_cb_lock); 4173 } 4174 EXPORT_SYMBOL(xfrm_if_register_cb); 4175 4176 void xfrm_if_unregister_cb(void) 4177 { 4178 RCU_INIT_POINTER(xfrm_if_cb, NULL); 4179 synchronize_rcu(); 4180 } 4181 EXPORT_SYMBOL(xfrm_if_unregister_cb); 4182 4183 #ifdef CONFIG_XFRM_STATISTICS 4184 static int __net_init xfrm_statistics_init(struct net *net) 4185 { 4186 int rv; 4187 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib); 4188 if (!net->mib.xfrm_statistics) 4189 return -ENOMEM; 4190 rv = xfrm_proc_init(net); 4191 if (rv < 0) 4192 free_percpu(net->mib.xfrm_statistics); 4193 return rv; 4194 } 4195 4196 static void xfrm_statistics_fini(struct net *net) 4197 { 4198 xfrm_proc_fini(net); 4199 free_percpu(net->mib.xfrm_statistics); 4200 } 4201 #else 4202 static int __net_init xfrm_statistics_init(struct net *net) 4203 { 4204 return 0; 4205 } 4206 4207 static void xfrm_statistics_fini(struct net *net) 4208 { 4209 } 4210 #endif 4211 4212 static int __net_init xfrm_policy_init(struct net *net) 4213 { 4214 unsigned int hmask, sz; 4215 int dir, err; 4216 4217 if (net_eq(net, &init_net)) { 4218 xfrm_dst_cache = KMEM_CACHE(xfrm_dst, SLAB_HWCACHE_ALIGN | SLAB_PANIC); 4219 err = rhashtable_init(&xfrm_policy_inexact_table, 4220 &xfrm_pol_inexact_params); 4221 BUG_ON(err); 4222 } 4223 4224 hmask = 8 - 1; 4225 sz = (hmask+1) * sizeof(struct hlist_head); 4226 4227 net->xfrm.policy_byidx = xfrm_hash_alloc(sz); 4228 if (!net->xfrm.policy_byidx) 4229 goto out_byidx; 4230 net->xfrm.policy_idx_hmask = hmask; 4231 4232 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 4233 struct xfrm_policy_hash *htab; 4234 4235 net->xfrm.policy_count[dir] = 0; 4236 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0; 4237 4238 htab = &net->xfrm.policy_bydst[dir]; 4239 rcu_assign_pointer(htab->table, xfrm_hash_alloc(sz)); 4240 if (!htab->table) 4241 goto out_bydst; 4242 htab->hmask = hmask; 4243 htab->dbits4 = 32; 4244 htab->sbits4 = 32; 4245 htab->dbits6 = 128; 4246 htab->sbits6 = 128; 4247 } 4248 net->xfrm.policy_hthresh.lbits4 = 32; 4249 net->xfrm.policy_hthresh.rbits4 = 32; 4250 net->xfrm.policy_hthresh.lbits6 = 128; 4251 net->xfrm.policy_hthresh.rbits6 = 128; 4252 4253 seqlock_init(&net->xfrm.policy_hthresh.lock); 4254 4255 INIT_LIST_HEAD(&net->xfrm.policy_all); 4256 INIT_LIST_HEAD(&net->xfrm.inexact_bins); 4257 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize); 4258 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild); 4259 return 0; 4260 4261 out_bydst: 4262 for (dir--; dir >= 0; dir--) { 4263 struct xfrm_policy_hash *htab; 4264 4265 htab = &net->xfrm.policy_bydst[dir]; 4266 xfrm_hash_free(rcu_dereference_protected(htab->table, true), sz); 4267 } 4268 xfrm_hash_free(net->xfrm.policy_byidx, sz); 4269 out_byidx: 4270 return -ENOMEM; 4271 } 4272 4273 static void __net_exit xfrm_net_pre_exit(struct net *net) 4274 { 4275 disable_work_sync(&net->xfrm.policy_hthresh.work); 4276 flush_work(&net->xfrm.policy_hash_work); 4277 #ifdef CONFIG_XFRM_SUB_POLICY 4278 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false); 4279 #endif 4280 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false); 4281 } 4282 4283 static void xfrm_policy_fini(struct net *net) 4284 { 4285 struct xfrm_pol_inexact_bin *b, *t; 4286 unsigned int sz; 4287 int dir; 4288 4289 WARN_ON(!list_empty(&net->xfrm.policy_all)); 4290 4291 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 4292 struct xfrm_policy_hash *htab; 4293 4294 htab = &net->xfrm.policy_bydst[dir]; 4295 sz = (htab->hmask + 1) * sizeof(struct hlist_head); 4296 WARN_ON(!hlist_empty(rcu_dereference_protected(htab->table, true))); 4297 xfrm_hash_free(rcu_dereference_protected(htab->table, true), sz); 4298 } 4299 4300 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head); 4301 WARN_ON(!hlist_empty(net->xfrm.policy_byidx)); 4302 xfrm_hash_free(net->xfrm.policy_byidx, sz); 4303 4304 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 4305 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins) 4306 __xfrm_policy_inexact_prune_bin(b, true); 4307 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 4308 } 4309 4310 static int __net_init xfrm_net_init(struct net *net) 4311 { 4312 int rv; 4313 4314 /* Initialize the per-net locks here */ 4315 spin_lock_init(&net->xfrm.xfrm_state_lock); 4316 spin_lock_init(&net->xfrm.xfrm_policy_lock); 4317 seqcount_spinlock_init(&net->xfrm.xfrm_policy_hash_generation, &net->xfrm.xfrm_policy_lock); 4318 mutex_init(&net->xfrm.xfrm_cfg_mutex); 4319 net->xfrm.policy_default[XFRM_POLICY_IN] = XFRM_USERPOLICY_ACCEPT; 4320 net->xfrm.policy_default[XFRM_POLICY_FWD] = XFRM_USERPOLICY_ACCEPT; 4321 net->xfrm.policy_default[XFRM_POLICY_OUT] = XFRM_USERPOLICY_ACCEPT; 4322 4323 rv = xfrm_statistics_init(net); 4324 if (rv < 0) 4325 goto out_statistics; 4326 rv = xfrm_state_init(net); 4327 if (rv < 0) 4328 goto out_state; 4329 rv = xfrm_policy_init(net); 4330 if (rv < 0) 4331 goto out_policy; 4332 rv = xfrm_sysctl_init(net); 4333 if (rv < 0) 4334 goto out_sysctl; 4335 4336 rv = xfrm_nat_keepalive_net_init(net); 4337 if (rv < 0) 4338 goto out_nat_keepalive; 4339 4340 return 0; 4341 4342 out_nat_keepalive: 4343 xfrm_sysctl_fini(net); 4344 out_sysctl: 4345 xfrm_policy_fini(net); 4346 out_policy: 4347 xfrm_state_fini(net); 4348 out_state: 4349 xfrm_statistics_fini(net); 4350 out_statistics: 4351 return rv; 4352 } 4353 4354 static void __net_exit xfrm_net_exit(struct net *net) 4355 { 4356 xfrm_nat_keepalive_net_fini(net); 4357 xfrm_sysctl_fini(net); 4358 xfrm_policy_fini(net); 4359 xfrm_state_fini(net); 4360 xfrm_statistics_fini(net); 4361 } 4362 4363 static struct pernet_operations __net_initdata xfrm_net_ops = { 4364 .init = xfrm_net_init, 4365 .pre_exit = xfrm_net_pre_exit, 4366 .exit = xfrm_net_exit, 4367 }; 4368 4369 static const struct flow_dissector_key xfrm_flow_dissector_keys[] = { 4370 { 4371 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 4372 .offset = offsetof(struct xfrm_flow_keys, control), 4373 }, 4374 { 4375 .key_id = FLOW_DISSECTOR_KEY_BASIC, 4376 .offset = offsetof(struct xfrm_flow_keys, basic), 4377 }, 4378 { 4379 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 4380 .offset = offsetof(struct xfrm_flow_keys, addrs.ipv4), 4381 }, 4382 { 4383 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 4384 .offset = offsetof(struct xfrm_flow_keys, addrs.ipv6), 4385 }, 4386 { 4387 .key_id = FLOW_DISSECTOR_KEY_PORTS, 4388 .offset = offsetof(struct xfrm_flow_keys, ports), 4389 }, 4390 { 4391 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 4392 .offset = offsetof(struct xfrm_flow_keys, gre), 4393 }, 4394 { 4395 .key_id = FLOW_DISSECTOR_KEY_IP, 4396 .offset = offsetof(struct xfrm_flow_keys, ip), 4397 }, 4398 { 4399 .key_id = FLOW_DISSECTOR_KEY_ICMP, 4400 .offset = offsetof(struct xfrm_flow_keys, icmp), 4401 }, 4402 }; 4403 4404 void __init xfrm_init(void) 4405 { 4406 skb_flow_dissector_init(&xfrm_session_dissector, 4407 xfrm_flow_dissector_keys, 4408 ARRAY_SIZE(xfrm_flow_dissector_keys)); 4409 4410 register_pernet_subsys(&xfrm_net_ops); 4411 xfrm_dev_init(); 4412 xfrm_input_init(); 4413 4414 #ifdef CONFIG_XFRM_ESPINTCP 4415 espintcp_init(); 4416 #endif 4417 4418 register_xfrm_state_bpf(); 4419 xfrm_nat_keepalive_init(AF_INET); 4420 } 4421 4422 #ifdef CONFIG_AUDITSYSCALL 4423 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp, 4424 struct audit_buffer *audit_buf) 4425 { 4426 struct xfrm_sec_ctx *ctx = xp->security; 4427 struct xfrm_selector *sel = &xp->selector; 4428 4429 if (ctx) 4430 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s", 4431 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str); 4432 4433 switch (sel->family) { 4434 case AF_INET: 4435 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4); 4436 if (sel->prefixlen_s != 32) 4437 audit_log_format(audit_buf, " src_prefixlen=%d", 4438 sel->prefixlen_s); 4439 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4); 4440 if (sel->prefixlen_d != 32) 4441 audit_log_format(audit_buf, " dst_prefixlen=%d", 4442 sel->prefixlen_d); 4443 break; 4444 case AF_INET6: 4445 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6); 4446 if (sel->prefixlen_s != 128) 4447 audit_log_format(audit_buf, " src_prefixlen=%d", 4448 sel->prefixlen_s); 4449 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6); 4450 if (sel->prefixlen_d != 128) 4451 audit_log_format(audit_buf, " dst_prefixlen=%d", 4452 sel->prefixlen_d); 4453 break; 4454 } 4455 } 4456 4457 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid) 4458 { 4459 struct audit_buffer *audit_buf; 4460 4461 audit_buf = xfrm_audit_start("SPD-add"); 4462 if (audit_buf == NULL) 4463 return; 4464 xfrm_audit_helper_usrinfo(task_valid, audit_buf); 4465 audit_log_format(audit_buf, " res=%u", result); 4466 xfrm_audit_common_policyinfo(xp, audit_buf); 4467 audit_log_end(audit_buf); 4468 } 4469 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add); 4470 4471 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result, 4472 bool task_valid) 4473 { 4474 struct audit_buffer *audit_buf; 4475 4476 audit_buf = xfrm_audit_start("SPD-delete"); 4477 if (audit_buf == NULL) 4478 return; 4479 xfrm_audit_helper_usrinfo(task_valid, audit_buf); 4480 audit_log_format(audit_buf, " res=%u", result); 4481 xfrm_audit_common_policyinfo(xp, audit_buf); 4482 audit_log_end(audit_buf); 4483 } 4484 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete); 4485 #endif 4486 4487 #ifdef CONFIG_XFRM_MIGRATE 4488 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel, 4489 u8 dir, u8 type, struct net *net, u32 if_id) 4490 { 4491 struct xfrm_policy *pol; 4492 struct flowi fl; 4493 4494 memset(&fl, 0, sizeof(fl)); 4495 4496 fl.flowi_proto = sel->proto; 4497 4498 switch (sel->family) { 4499 case AF_INET: 4500 fl.u.ip4.saddr = sel->saddr.a4; 4501 fl.u.ip4.daddr = sel->daddr.a4; 4502 if (sel->proto == IPSEC_ULPROTO_ANY) 4503 break; 4504 fl.u.flowi4_oif = sel->ifindex; 4505 fl.u.ip4.fl4_sport = sel->sport; 4506 fl.u.ip4.fl4_dport = sel->dport; 4507 break; 4508 case AF_INET6: 4509 fl.u.ip6.saddr = sel->saddr.in6; 4510 fl.u.ip6.daddr = sel->daddr.in6; 4511 if (sel->proto == IPSEC_ULPROTO_ANY) 4512 break; 4513 fl.u.flowi6_oif = sel->ifindex; 4514 fl.u.ip6.fl4_sport = sel->sport; 4515 fl.u.ip6.fl4_dport = sel->dport; 4516 break; 4517 default: 4518 return ERR_PTR(-EAFNOSUPPORT); 4519 } 4520 4521 rcu_read_lock(); 4522 4523 pol = xfrm_policy_lookup_bytype(net, type, &fl, sel->family, dir, if_id); 4524 if (IS_ERR_OR_NULL(pol)) 4525 goto out_unlock; 4526 out_unlock: 4527 rcu_read_unlock(); 4528 return pol; 4529 } 4530 4531 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t) 4532 { 4533 int match = 0; 4534 4535 if (t->mode == m->mode && t->id.proto == m->proto && 4536 (m->old_reqid == 0 || t->reqid == m->old_reqid)) { 4537 switch (t->mode) { 4538 case XFRM_MODE_TUNNEL: 4539 case XFRM_MODE_BEET: 4540 case XFRM_MODE_IPTFS: 4541 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr, 4542 m->old_family) && 4543 xfrm_addr_equal(&t->saddr, &m->old_saddr, 4544 m->old_family)) { 4545 match = 1; 4546 } 4547 break; 4548 case XFRM_MODE_TRANSPORT: 4549 /* in case of transport mode, template does not store 4550 any IP addresses, hence we just compare mode and 4551 protocol */ 4552 match = 1; 4553 break; 4554 default: 4555 break; 4556 } 4557 } 4558 return match; 4559 } 4560 4561 /* update endpoint address(es) of template(s) */ 4562 static int xfrm_policy_migrate(struct xfrm_policy *pol, 4563 struct xfrm_migrate *m, int num_migrate, 4564 struct netlink_ext_ack *extack) 4565 { 4566 struct xfrm_migrate *mp; 4567 int i, j, n = 0; 4568 4569 write_lock_bh(&pol->lock); 4570 if (unlikely(pol->walk.dead)) { 4571 /* target policy has been deleted */ 4572 NL_SET_ERR_MSG(extack, "Target policy not found"); 4573 write_unlock_bh(&pol->lock); 4574 return -ENOENT; 4575 } 4576 4577 for (i = 0; i < pol->xfrm_nr; i++) { 4578 for (j = 0, mp = m; j < num_migrate; j++, mp++) { 4579 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i])) 4580 continue; 4581 n++; 4582 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL && 4583 pol->xfrm_vec[i].mode != XFRM_MODE_BEET && 4584 pol->xfrm_vec[i].mode != XFRM_MODE_IPTFS) 4585 continue; 4586 /* update endpoints */ 4587 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr, 4588 sizeof(pol->xfrm_vec[i].id.daddr)); 4589 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr, 4590 sizeof(pol->xfrm_vec[i].saddr)); 4591 pol->xfrm_vec[i].encap_family = mp->new_family; 4592 /* flush bundles */ 4593 atomic_inc(&pol->genid); 4594 } 4595 } 4596 4597 write_unlock_bh(&pol->lock); 4598 4599 if (!n) 4600 return -ENODATA; 4601 4602 return 0; 4603 } 4604 4605 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate, 4606 struct netlink_ext_ack *extack) 4607 { 4608 int i, j; 4609 4610 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH) { 4611 NL_SET_ERR_MSG(extack, "Invalid number of SAs to migrate, must be 0 < num <= XFRM_MAX_DEPTH (6)"); 4612 return -EINVAL; 4613 } 4614 4615 for (i = 0; i < num_migrate; i++) { 4616 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) || 4617 xfrm_addr_any(&m[i].new_saddr, m[i].new_family)) { 4618 NL_SET_ERR_MSG(extack, "Addresses in the MIGRATE attribute's list cannot be null"); 4619 return -EINVAL; 4620 } 4621 4622 /* check if there is any duplicated entry */ 4623 for (j = i + 1; j < num_migrate; j++) { 4624 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr, 4625 sizeof(m[i].old_daddr)) && 4626 !memcmp(&m[i].old_saddr, &m[j].old_saddr, 4627 sizeof(m[i].old_saddr)) && 4628 m[i].proto == m[j].proto && 4629 m[i].mode == m[j].mode && 4630 m[i].old_reqid == m[j].old_reqid && 4631 m[i].old_family == m[j].old_family) { 4632 NL_SET_ERR_MSG(extack, "Entries in the MIGRATE attribute's list must be unique"); 4633 return -EINVAL; 4634 } 4635 } 4636 } 4637 4638 return 0; 4639 } 4640 4641 /* 4642 * Fill migrate fields that are invariant in XFRM_MSG_MIGRATE: inherited 4643 * from the existing SA unchanged. XFRM_MSG_MIGRATE_STATE can update these. 4644 */ 4645 static void xfrm_migrate_copy_old(const struct xfrm_state *x, 4646 struct xfrm_migrate *mp) 4647 { 4648 mp->msg_type = XFRM_MSG_MIGRATE; 4649 mp->smark = x->props.smark; 4650 mp->new_reqid = x->props.reqid; 4651 mp->nat_keepalive_interval = x->nat_keepalive_interval; 4652 mp->mapping_maxage = x->mapping_maxage; 4653 mp->new_mark = &x->mark; 4654 } 4655 4656 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type, 4657 struct xfrm_migrate *m, int num_migrate, 4658 struct xfrm_kmaddress *k, struct net *net, 4659 struct xfrm_encap_tmpl *encap, u32 if_id, 4660 struct netlink_ext_ack *extack, struct xfrm_user_offload *xuo) 4661 { 4662 int i, err, nx_cur = 0, nx_new = 0; 4663 struct xfrm_policy *pol = NULL; 4664 struct xfrm_state *x, *xc; 4665 struct xfrm_state *x_cur[XFRM_MAX_DEPTH]; 4666 struct xfrm_state *x_new[XFRM_MAX_DEPTH]; 4667 struct xfrm_migrate *mp; 4668 4669 /* Stage 0 - sanity checks */ 4670 err = xfrm_migrate_check(m, num_migrate, extack); 4671 if (err < 0) 4672 goto out; 4673 4674 if (dir >= XFRM_POLICY_MAX) { 4675 NL_SET_ERR_MSG(extack, "Invalid policy direction"); 4676 err = -EINVAL; 4677 goto out; 4678 } 4679 4680 /* Stage 1 - find policy */ 4681 pol = xfrm_migrate_policy_find(sel, dir, type, net, if_id); 4682 if (IS_ERR_OR_NULL(pol)) { 4683 NL_SET_ERR_MSG(extack, "Target policy not found"); 4684 err = IS_ERR(pol) ? PTR_ERR(pol) : -ENOENT; 4685 goto out; 4686 } 4687 4688 /* Stage 2 - find and update state(s) */ 4689 for (i = 0, mp = m; i < num_migrate; i++, mp++) { 4690 if ((x = xfrm_migrate_state_find(mp, net, if_id))) { 4691 x_cur[nx_cur] = x; 4692 nx_cur++; 4693 mp->encap = encap; 4694 mp->xuo = xuo; 4695 xfrm_migrate_copy_old(x, mp); 4696 4697 xc = xfrm_state_migrate(x, mp, net, extack); 4698 if (xc) { 4699 x_new[nx_new] = xc; 4700 nx_new++; 4701 } else { 4702 err = -ENODATA; 4703 goto restore_state; 4704 } 4705 } 4706 } 4707 4708 /* Stage 3 - update policy */ 4709 err = xfrm_policy_migrate(pol, m, num_migrate, extack); 4710 if (err < 0) 4711 goto restore_state; 4712 4713 /* Stage 4 - delete old state(s) */ 4714 if (nx_cur) { 4715 xfrm_states_put(x_cur, nx_cur); 4716 xfrm_states_delete(x_cur, nx_cur); 4717 } 4718 4719 /* Stage 5 - announce */ 4720 km_migrate(sel, dir, type, m, num_migrate, k, net, encap); 4721 4722 xfrm_pol_put(pol); 4723 4724 return 0; 4725 out: 4726 return err; 4727 4728 restore_state: 4729 if (pol) 4730 xfrm_pol_put(pol); 4731 if (nx_cur) 4732 xfrm_states_put(x_cur, nx_cur); 4733 if (nx_new) 4734 xfrm_states_delete(x_new, nx_new); 4735 4736 return err; 4737 } 4738 EXPORT_SYMBOL(xfrm_migrate); 4739 #endif 4740