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