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