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/xfrm.h> 35 #include <net/ip.h> 36 #if IS_ENABLED(CONFIG_IPV6_MIP6) 37 #include <net/mip6.h> 38 #endif 39 #ifdef CONFIG_XFRM_STATISTICS 40 #include <net/snmp.h> 41 #endif 42 43 #include "xfrm_hash.h" 44 45 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10)) 46 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ)) 47 #define XFRM_MAX_QUEUE_LEN 100 48 49 struct xfrm_flo { 50 struct dst_entry *dst_orig; 51 u8 flags; 52 }; 53 54 /* prefixes smaller than this are stored in lists, not trees. */ 55 #define INEXACT_PREFIXLEN_IPV4 16 56 #define INEXACT_PREFIXLEN_IPV6 48 57 58 struct xfrm_pol_inexact_node { 59 struct rb_node node; 60 union { 61 xfrm_address_t addr; 62 struct rcu_head rcu; 63 }; 64 u8 prefixlen; 65 66 struct rb_root root; 67 68 /* the policies matching this node, can be empty list */ 69 struct hlist_head hhead; 70 }; 71 72 /* xfrm inexact policy search tree: 73 * xfrm_pol_inexact_bin = hash(dir,type,family,if_id); 74 * | 75 * +---- root_d: sorted by daddr:prefix 76 * | | 77 * | xfrm_pol_inexact_node 78 * | | 79 * | +- root: sorted by saddr/prefix 80 * | | | 81 * | | xfrm_pol_inexact_node 82 * | | | 83 * | | + root: unused 84 * | | | 85 * | | + hhead: saddr:daddr policies 86 * | | 87 * | +- coarse policies and all any:daddr policies 88 * | 89 * +---- root_s: sorted by saddr:prefix 90 * | | 91 * | xfrm_pol_inexact_node 92 * | | 93 * | + root: unused 94 * | | 95 * | + hhead: saddr:any policies 96 * | 97 * +---- coarse policies and all any:any policies 98 * 99 * Lookups return four candidate lists: 100 * 1. any:any list from top-level xfrm_pol_inexact_bin 101 * 2. any:daddr list from daddr tree 102 * 3. saddr:daddr list from 2nd level daddr tree 103 * 4. saddr:any list from saddr tree 104 * 105 * This result set then needs to be searched for the policy with 106 * the lowest priority. If two results have same prio, youngest one wins. 107 */ 108 109 struct xfrm_pol_inexact_key { 110 possible_net_t net; 111 u32 if_id; 112 u16 family; 113 u8 dir, type; 114 }; 115 116 struct xfrm_pol_inexact_bin { 117 struct xfrm_pol_inexact_key k; 118 struct rhash_head head; 119 /* list containing '*:*' policies */ 120 struct hlist_head hhead; 121 122 seqcount_t count; 123 /* tree sorted by daddr/prefix */ 124 struct rb_root root_d; 125 126 /* tree sorted by saddr/prefix */ 127 struct rb_root root_s; 128 129 /* slow path below */ 130 struct list_head inexact_bins; 131 struct rcu_head rcu; 132 }; 133 134 enum xfrm_pol_inexact_candidate_type { 135 XFRM_POL_CAND_BOTH, 136 XFRM_POL_CAND_SADDR, 137 XFRM_POL_CAND_DADDR, 138 XFRM_POL_CAND_ANY, 139 140 XFRM_POL_CAND_MAX, 141 }; 142 143 struct xfrm_pol_inexact_candidates { 144 struct hlist_head *res[XFRM_POL_CAND_MAX]; 145 }; 146 147 static DEFINE_SPINLOCK(xfrm_if_cb_lock); 148 static struct xfrm_if_cb const __rcu *xfrm_if_cb __read_mostly; 149 150 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock); 151 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1] 152 __read_mostly; 153 154 static struct kmem_cache *xfrm_dst_cache __ro_after_init; 155 static __read_mostly seqcount_t xfrm_policy_hash_generation; 156 157 static struct rhashtable xfrm_policy_inexact_table; 158 static const struct rhashtable_params xfrm_pol_inexact_params; 159 160 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr); 161 static int stale_bundle(struct dst_entry *dst); 162 static int xfrm_bundle_ok(struct xfrm_dst *xdst); 163 static void xfrm_policy_queue_process(struct timer_list *t); 164 165 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir); 166 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol, 167 int dir); 168 169 static struct xfrm_pol_inexact_bin * 170 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, u8 dir, 171 u32 if_id); 172 173 static struct xfrm_pol_inexact_bin * 174 xfrm_policy_inexact_lookup_rcu(struct net *net, 175 u8 type, u16 family, u8 dir, u32 if_id); 176 static struct xfrm_policy * 177 xfrm_policy_insert_list(struct hlist_head *chain, struct xfrm_policy *policy, 178 bool excl); 179 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain, 180 struct xfrm_policy *policy); 181 182 static bool 183 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand, 184 struct xfrm_pol_inexact_bin *b, 185 const xfrm_address_t *saddr, 186 const xfrm_address_t *daddr); 187 188 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy) 189 { 190 return refcount_inc_not_zero(&policy->refcnt); 191 } 192 193 static inline bool 194 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl) 195 { 196 const struct flowi4 *fl4 = &fl->u.ip4; 197 198 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) && 199 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) && 200 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) && 201 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) && 202 (fl4->flowi4_proto == sel->proto || !sel->proto) && 203 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex); 204 } 205 206 static inline bool 207 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl) 208 { 209 const struct flowi6 *fl6 = &fl->u.ip6; 210 211 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) && 212 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) && 213 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) && 214 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) && 215 (fl6->flowi6_proto == sel->proto || !sel->proto) && 216 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex); 217 } 218 219 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl, 220 unsigned short family) 221 { 222 switch (family) { 223 case AF_INET: 224 return __xfrm4_selector_match(sel, fl); 225 case AF_INET6: 226 return __xfrm6_selector_match(sel, fl); 227 } 228 return false; 229 } 230 231 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family) 232 { 233 const struct xfrm_policy_afinfo *afinfo; 234 235 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo))) 236 return NULL; 237 rcu_read_lock(); 238 afinfo = rcu_dereference(xfrm_policy_afinfo[family]); 239 if (unlikely(!afinfo)) 240 rcu_read_unlock(); 241 return afinfo; 242 } 243 244 /* Called with rcu_read_lock(). */ 245 static const struct xfrm_if_cb *xfrm_if_get_cb(void) 246 { 247 return rcu_dereference(xfrm_if_cb); 248 } 249 250 struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos, int oif, 251 const xfrm_address_t *saddr, 252 const xfrm_address_t *daddr, 253 int family, u32 mark) 254 { 255 const struct xfrm_policy_afinfo *afinfo; 256 struct dst_entry *dst; 257 258 afinfo = xfrm_policy_get_afinfo(family); 259 if (unlikely(afinfo == NULL)) 260 return ERR_PTR(-EAFNOSUPPORT); 261 262 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr, mark); 263 264 rcu_read_unlock(); 265 266 return dst; 267 } 268 EXPORT_SYMBOL(__xfrm_dst_lookup); 269 270 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, 271 int tos, int oif, 272 xfrm_address_t *prev_saddr, 273 xfrm_address_t *prev_daddr, 274 int family, u32 mark) 275 { 276 struct net *net = xs_net(x); 277 xfrm_address_t *saddr = &x->props.saddr; 278 xfrm_address_t *daddr = &x->id.daddr; 279 struct dst_entry *dst; 280 281 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) { 282 saddr = x->coaddr; 283 daddr = prev_daddr; 284 } 285 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) { 286 saddr = prev_saddr; 287 daddr = x->coaddr; 288 } 289 290 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family, mark); 291 292 if (!IS_ERR(dst)) { 293 if (prev_saddr != saddr) 294 memcpy(prev_saddr, saddr, sizeof(*prev_saddr)); 295 if (prev_daddr != daddr) 296 memcpy(prev_daddr, daddr, sizeof(*prev_daddr)); 297 } 298 299 return dst; 300 } 301 302 static inline unsigned long make_jiffies(long secs) 303 { 304 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ) 305 return MAX_SCHEDULE_TIMEOUT-1; 306 else 307 return secs*HZ; 308 } 309 310 static void xfrm_policy_timer(struct timer_list *t) 311 { 312 struct xfrm_policy *xp = from_timer(xp, t, timer); 313 time64_t now = ktime_get_real_seconds(); 314 time64_t next = TIME64_MAX; 315 int warn = 0; 316 int dir; 317 318 read_lock(&xp->lock); 319 320 if (unlikely(xp->walk.dead)) 321 goto out; 322 323 dir = xfrm_policy_id2dir(xp->index); 324 325 if (xp->lft.hard_add_expires_seconds) { 326 time64_t tmo = xp->lft.hard_add_expires_seconds + 327 xp->curlft.add_time - now; 328 if (tmo <= 0) 329 goto expired; 330 if (tmo < next) 331 next = tmo; 332 } 333 if (xp->lft.hard_use_expires_seconds) { 334 time64_t tmo = xp->lft.hard_use_expires_seconds + 335 (xp->curlft.use_time ? : xp->curlft.add_time) - now; 336 if (tmo <= 0) 337 goto expired; 338 if (tmo < next) 339 next = tmo; 340 } 341 if (xp->lft.soft_add_expires_seconds) { 342 time64_t tmo = xp->lft.soft_add_expires_seconds + 343 xp->curlft.add_time - now; 344 if (tmo <= 0) { 345 warn = 1; 346 tmo = XFRM_KM_TIMEOUT; 347 } 348 if (tmo < next) 349 next = tmo; 350 } 351 if (xp->lft.soft_use_expires_seconds) { 352 time64_t tmo = xp->lft.soft_use_expires_seconds + 353 (xp->curlft.use_time ? : xp->curlft.add_time) - now; 354 if (tmo <= 0) { 355 warn = 1; 356 tmo = XFRM_KM_TIMEOUT; 357 } 358 if (tmo < next) 359 next = tmo; 360 } 361 362 if (warn) 363 km_policy_expired(xp, dir, 0, 0); 364 if (next != TIME64_MAX && 365 !mod_timer(&xp->timer, jiffies + make_jiffies(next))) 366 xfrm_pol_hold(xp); 367 368 out: 369 read_unlock(&xp->lock); 370 xfrm_pol_put(xp); 371 return; 372 373 expired: 374 read_unlock(&xp->lock); 375 if (!xfrm_policy_delete(xp, dir)) 376 km_policy_expired(xp, dir, 1, 0); 377 xfrm_pol_put(xp); 378 } 379 380 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2 381 * SPD calls. 382 */ 383 384 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp) 385 { 386 struct xfrm_policy *policy; 387 388 policy = kzalloc(sizeof(struct xfrm_policy), gfp); 389 390 if (policy) { 391 write_pnet(&policy->xp_net, net); 392 INIT_LIST_HEAD(&policy->walk.all); 393 INIT_HLIST_NODE(&policy->bydst_inexact_list); 394 INIT_HLIST_NODE(&policy->bydst); 395 INIT_HLIST_NODE(&policy->byidx); 396 rwlock_init(&policy->lock); 397 refcount_set(&policy->refcnt, 1); 398 skb_queue_head_init(&policy->polq.hold_queue); 399 timer_setup(&policy->timer, xfrm_policy_timer, 0); 400 timer_setup(&policy->polq.hold_timer, 401 xfrm_policy_queue_process, 0); 402 } 403 return policy; 404 } 405 EXPORT_SYMBOL(xfrm_policy_alloc); 406 407 static void xfrm_policy_destroy_rcu(struct rcu_head *head) 408 { 409 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu); 410 411 security_xfrm_policy_free(policy->security); 412 kfree(policy); 413 } 414 415 /* Destroy xfrm_policy: descendant resources must be released to this moment. */ 416 417 void xfrm_policy_destroy(struct xfrm_policy *policy) 418 { 419 BUG_ON(!policy->walk.dead); 420 421 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer)) 422 BUG(); 423 424 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu); 425 } 426 EXPORT_SYMBOL(xfrm_policy_destroy); 427 428 /* Rule must be locked. Release descendant resources, announce 429 * entry dead. The rule must be unlinked from lists to the moment. 430 */ 431 432 static void xfrm_policy_kill(struct xfrm_policy *policy) 433 { 434 policy->walk.dead = 1; 435 436 atomic_inc(&policy->genid); 437 438 if (del_timer(&policy->polq.hold_timer)) 439 xfrm_pol_put(policy); 440 skb_queue_purge(&policy->polq.hold_queue); 441 442 if (del_timer(&policy->timer)) 443 xfrm_pol_put(policy); 444 445 xfrm_pol_put(policy); 446 } 447 448 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024; 449 450 static inline unsigned int idx_hash(struct net *net, u32 index) 451 { 452 return __idx_hash(index, net->xfrm.policy_idx_hmask); 453 } 454 455 /* calculate policy hash thresholds */ 456 static void __get_hash_thresh(struct net *net, 457 unsigned short family, int dir, 458 u8 *dbits, u8 *sbits) 459 { 460 switch (family) { 461 case AF_INET: 462 *dbits = net->xfrm.policy_bydst[dir].dbits4; 463 *sbits = net->xfrm.policy_bydst[dir].sbits4; 464 break; 465 466 case AF_INET6: 467 *dbits = net->xfrm.policy_bydst[dir].dbits6; 468 *sbits = net->xfrm.policy_bydst[dir].sbits6; 469 break; 470 471 default: 472 *dbits = 0; 473 *sbits = 0; 474 } 475 } 476 477 static struct hlist_head *policy_hash_bysel(struct net *net, 478 const struct xfrm_selector *sel, 479 unsigned short family, int dir) 480 { 481 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 482 unsigned int hash; 483 u8 dbits; 484 u8 sbits; 485 486 __get_hash_thresh(net, family, dir, &dbits, &sbits); 487 hash = __sel_hash(sel, family, hmask, dbits, sbits); 488 489 if (hash == hmask + 1) 490 return NULL; 491 492 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table, 493 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash; 494 } 495 496 static struct hlist_head *policy_hash_direct(struct net *net, 497 const xfrm_address_t *daddr, 498 const xfrm_address_t *saddr, 499 unsigned short family, int dir) 500 { 501 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 502 unsigned int hash; 503 u8 dbits; 504 u8 sbits; 505 506 __get_hash_thresh(net, family, dir, &dbits, &sbits); 507 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits); 508 509 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table, 510 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash; 511 } 512 513 static void xfrm_dst_hash_transfer(struct net *net, 514 struct hlist_head *list, 515 struct hlist_head *ndsttable, 516 unsigned int nhashmask, 517 int dir) 518 { 519 struct hlist_node *tmp, *entry0 = NULL; 520 struct xfrm_policy *pol; 521 unsigned int h0 = 0; 522 u8 dbits; 523 u8 sbits; 524 525 redo: 526 hlist_for_each_entry_safe(pol, tmp, list, bydst) { 527 unsigned int h; 528 529 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits); 530 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr, 531 pol->family, nhashmask, dbits, sbits); 532 if (!entry0) { 533 hlist_del_rcu(&pol->bydst); 534 hlist_add_head_rcu(&pol->bydst, ndsttable + h); 535 h0 = h; 536 } else { 537 if (h != h0) 538 continue; 539 hlist_del_rcu(&pol->bydst); 540 hlist_add_behind_rcu(&pol->bydst, entry0); 541 } 542 entry0 = &pol->bydst; 543 } 544 if (!hlist_empty(list)) { 545 entry0 = NULL; 546 goto redo; 547 } 548 } 549 550 static void xfrm_idx_hash_transfer(struct hlist_head *list, 551 struct hlist_head *nidxtable, 552 unsigned int nhashmask) 553 { 554 struct hlist_node *tmp; 555 struct xfrm_policy *pol; 556 557 hlist_for_each_entry_safe(pol, tmp, list, byidx) { 558 unsigned int h; 559 560 h = __idx_hash(pol->index, nhashmask); 561 hlist_add_head(&pol->byidx, nidxtable+h); 562 } 563 } 564 565 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask) 566 { 567 return ((old_hmask + 1) << 1) - 1; 568 } 569 570 static void xfrm_bydst_resize(struct net *net, int dir) 571 { 572 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 573 unsigned int nhashmask = xfrm_new_hash_mask(hmask); 574 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head); 575 struct hlist_head *ndst = xfrm_hash_alloc(nsize); 576 struct hlist_head *odst; 577 int i; 578 579 if (!ndst) 580 return; 581 582 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 583 write_seqcount_begin(&xfrm_policy_hash_generation); 584 585 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table, 586 lockdep_is_held(&net->xfrm.xfrm_policy_lock)); 587 588 for (i = hmask; i >= 0; i--) 589 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir); 590 591 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst); 592 net->xfrm.policy_bydst[dir].hmask = nhashmask; 593 594 write_seqcount_end(&xfrm_policy_hash_generation); 595 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 596 597 synchronize_rcu(); 598 599 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head)); 600 } 601 602 static void xfrm_byidx_resize(struct net *net, int total) 603 { 604 unsigned int hmask = net->xfrm.policy_idx_hmask; 605 unsigned int nhashmask = xfrm_new_hash_mask(hmask); 606 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head); 607 struct hlist_head *oidx = net->xfrm.policy_byidx; 608 struct hlist_head *nidx = xfrm_hash_alloc(nsize); 609 int i; 610 611 if (!nidx) 612 return; 613 614 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 615 616 for (i = hmask; i >= 0; i--) 617 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask); 618 619 net->xfrm.policy_byidx = nidx; 620 net->xfrm.policy_idx_hmask = nhashmask; 621 622 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 623 624 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head)); 625 } 626 627 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total) 628 { 629 unsigned int cnt = net->xfrm.policy_count[dir]; 630 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 631 632 if (total) 633 *total += cnt; 634 635 if ((hmask + 1) < xfrm_policy_hashmax && 636 cnt > hmask) 637 return 1; 638 639 return 0; 640 } 641 642 static inline int xfrm_byidx_should_resize(struct net *net, int total) 643 { 644 unsigned int hmask = net->xfrm.policy_idx_hmask; 645 646 if ((hmask + 1) < xfrm_policy_hashmax && 647 total > hmask) 648 return 1; 649 650 return 0; 651 } 652 653 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si) 654 { 655 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN]; 656 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT]; 657 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD]; 658 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX]; 659 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX]; 660 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX]; 661 si->spdhcnt = net->xfrm.policy_idx_hmask; 662 si->spdhmcnt = xfrm_policy_hashmax; 663 } 664 EXPORT_SYMBOL(xfrm_spd_getinfo); 665 666 static DEFINE_MUTEX(hash_resize_mutex); 667 static void xfrm_hash_resize(struct work_struct *work) 668 { 669 struct net *net = container_of(work, struct net, xfrm.policy_hash_work); 670 int dir, total; 671 672 mutex_lock(&hash_resize_mutex); 673 674 total = 0; 675 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 676 if (xfrm_bydst_should_resize(net, dir, &total)) 677 xfrm_bydst_resize(net, dir); 678 } 679 if (xfrm_byidx_should_resize(net, total)) 680 xfrm_byidx_resize(net, total); 681 682 mutex_unlock(&hash_resize_mutex); 683 } 684 685 /* Make sure *pol can be inserted into fastbin. 686 * Useful to check that later insert requests will be sucessful 687 * (provided xfrm_policy_lock is held throughout). 688 */ 689 static struct xfrm_pol_inexact_bin * 690 xfrm_policy_inexact_alloc_bin(const struct xfrm_policy *pol, u8 dir) 691 { 692 struct xfrm_pol_inexact_bin *bin, *prev; 693 struct xfrm_pol_inexact_key k = { 694 .family = pol->family, 695 .type = pol->type, 696 .dir = dir, 697 .if_id = pol->if_id, 698 }; 699 struct net *net = xp_net(pol); 700 701 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 702 703 write_pnet(&k.net, net); 704 bin = rhashtable_lookup_fast(&xfrm_policy_inexact_table, &k, 705 xfrm_pol_inexact_params); 706 if (bin) 707 return bin; 708 709 bin = kzalloc(sizeof(*bin), GFP_ATOMIC); 710 if (!bin) 711 return NULL; 712 713 bin->k = k; 714 INIT_HLIST_HEAD(&bin->hhead); 715 bin->root_d = RB_ROOT; 716 bin->root_s = RB_ROOT; 717 seqcount_init(&bin->count); 718 719 prev = rhashtable_lookup_get_insert_key(&xfrm_policy_inexact_table, 720 &bin->k, &bin->head, 721 xfrm_pol_inexact_params); 722 if (!prev) { 723 list_add(&bin->inexact_bins, &net->xfrm.inexact_bins); 724 return bin; 725 } 726 727 kfree(bin); 728 729 return IS_ERR(prev) ? NULL : prev; 730 } 731 732 static bool xfrm_pol_inexact_addr_use_any_list(const xfrm_address_t *addr, 733 int family, u8 prefixlen) 734 { 735 if (xfrm_addr_any(addr, family)) 736 return true; 737 738 if (family == AF_INET6 && prefixlen < INEXACT_PREFIXLEN_IPV6) 739 return true; 740 741 if (family == AF_INET && prefixlen < INEXACT_PREFIXLEN_IPV4) 742 return true; 743 744 return false; 745 } 746 747 static bool 748 xfrm_policy_inexact_insert_use_any_list(const struct xfrm_policy *policy) 749 { 750 const xfrm_address_t *addr; 751 bool saddr_any, daddr_any; 752 u8 prefixlen; 753 754 addr = &policy->selector.saddr; 755 prefixlen = policy->selector.prefixlen_s; 756 757 saddr_any = xfrm_pol_inexact_addr_use_any_list(addr, 758 policy->family, 759 prefixlen); 760 addr = &policy->selector.daddr; 761 prefixlen = policy->selector.prefixlen_d; 762 daddr_any = xfrm_pol_inexact_addr_use_any_list(addr, 763 policy->family, 764 prefixlen); 765 return saddr_any && daddr_any; 766 } 767 768 static void xfrm_pol_inexact_node_init(struct xfrm_pol_inexact_node *node, 769 const xfrm_address_t *addr, u8 prefixlen) 770 { 771 node->addr = *addr; 772 node->prefixlen = prefixlen; 773 } 774 775 static struct xfrm_pol_inexact_node * 776 xfrm_pol_inexact_node_alloc(const xfrm_address_t *addr, u8 prefixlen) 777 { 778 struct xfrm_pol_inexact_node *node; 779 780 node = kzalloc(sizeof(*node), GFP_ATOMIC); 781 if (node) 782 xfrm_pol_inexact_node_init(node, addr, prefixlen); 783 784 return node; 785 } 786 787 static int xfrm_policy_addr_delta(const xfrm_address_t *a, 788 const xfrm_address_t *b, 789 u8 prefixlen, u16 family) 790 { 791 unsigned int pdw, pbi; 792 int delta = 0; 793 794 switch (family) { 795 case AF_INET: 796 if (sizeof(long) == 4 && prefixlen == 0) 797 return ntohl(a->a4) - ntohl(b->a4); 798 return (ntohl(a->a4) & ((~0UL << (32 - prefixlen)))) - 799 (ntohl(b->a4) & ((~0UL << (32 - prefixlen)))); 800 case AF_INET6: 801 pdw = prefixlen >> 5; 802 pbi = prefixlen & 0x1f; 803 804 if (pdw) { 805 delta = memcmp(a->a6, b->a6, pdw << 2); 806 if (delta) 807 return delta; 808 } 809 if (pbi) { 810 u32 mask = ~0u << (32 - pbi); 811 812 delta = (ntohl(a->a6[pdw]) & mask) - 813 (ntohl(b->a6[pdw]) & mask); 814 } 815 break; 816 default: 817 break; 818 } 819 820 return delta; 821 } 822 823 static void xfrm_policy_inexact_list_reinsert(struct net *net, 824 struct xfrm_pol_inexact_node *n, 825 u16 family) 826 { 827 unsigned int matched_s, matched_d; 828 struct xfrm_policy *policy, *p; 829 830 matched_s = 0; 831 matched_d = 0; 832 833 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) { 834 struct hlist_node *newpos = NULL; 835 bool matches_s, matches_d; 836 837 if (!policy->bydst_reinsert) 838 continue; 839 840 WARN_ON_ONCE(policy->family != family); 841 842 policy->bydst_reinsert = false; 843 hlist_for_each_entry(p, &n->hhead, bydst) { 844 if (policy->priority > p->priority) 845 newpos = &p->bydst; 846 else if (policy->priority == p->priority && 847 policy->pos > p->pos) 848 newpos = &p->bydst; 849 else 850 break; 851 } 852 853 if (newpos) 854 hlist_add_behind_rcu(&policy->bydst, newpos); 855 else 856 hlist_add_head_rcu(&policy->bydst, &n->hhead); 857 858 /* paranoia checks follow. 859 * Check that the reinserted policy matches at least 860 * saddr or daddr for current node prefix. 861 * 862 * Matching both is fine, matching saddr in one policy 863 * (but not daddr) and then matching only daddr in another 864 * is a bug. 865 */ 866 matches_s = xfrm_policy_addr_delta(&policy->selector.saddr, 867 &n->addr, 868 n->prefixlen, 869 family) == 0; 870 matches_d = xfrm_policy_addr_delta(&policy->selector.daddr, 871 &n->addr, 872 n->prefixlen, 873 family) == 0; 874 if (matches_s && matches_d) 875 continue; 876 877 WARN_ON_ONCE(!matches_s && !matches_d); 878 if (matches_s) 879 matched_s++; 880 if (matches_d) 881 matched_d++; 882 WARN_ON_ONCE(matched_s && matched_d); 883 } 884 } 885 886 static void xfrm_policy_inexact_node_reinsert(struct net *net, 887 struct xfrm_pol_inexact_node *n, 888 struct rb_root *new, 889 u16 family) 890 { 891 struct xfrm_pol_inexact_node *node; 892 struct rb_node **p, *parent; 893 894 /* we should not have another subtree here */ 895 WARN_ON_ONCE(!RB_EMPTY_ROOT(&n->root)); 896 restart: 897 parent = NULL; 898 p = &new->rb_node; 899 while (*p) { 900 u8 prefixlen; 901 int delta; 902 903 parent = *p; 904 node = rb_entry(*p, struct xfrm_pol_inexact_node, node); 905 906 prefixlen = min(node->prefixlen, n->prefixlen); 907 908 delta = xfrm_policy_addr_delta(&n->addr, &node->addr, 909 prefixlen, family); 910 if (delta < 0) { 911 p = &parent->rb_left; 912 } else if (delta > 0) { 913 p = &parent->rb_right; 914 } else { 915 bool same_prefixlen = node->prefixlen == n->prefixlen; 916 struct xfrm_policy *tmp; 917 918 hlist_for_each_entry(tmp, &n->hhead, bydst) { 919 tmp->bydst_reinsert = true; 920 hlist_del_rcu(&tmp->bydst); 921 } 922 923 node->prefixlen = prefixlen; 924 925 xfrm_policy_inexact_list_reinsert(net, node, family); 926 927 if (same_prefixlen) { 928 kfree_rcu(n, rcu); 929 return; 930 } 931 932 rb_erase(*p, new); 933 kfree_rcu(n, rcu); 934 n = node; 935 goto restart; 936 } 937 } 938 939 rb_link_node_rcu(&n->node, parent, p); 940 rb_insert_color(&n->node, new); 941 } 942 943 /* merge nodes v and n */ 944 static void xfrm_policy_inexact_node_merge(struct net *net, 945 struct xfrm_pol_inexact_node *v, 946 struct xfrm_pol_inexact_node *n, 947 u16 family) 948 { 949 struct xfrm_pol_inexact_node *node; 950 struct xfrm_policy *tmp; 951 struct rb_node *rnode; 952 953 /* To-be-merged node v has a subtree. 954 * 955 * Dismantle it and insert its nodes to n->root. 956 */ 957 while ((rnode = rb_first(&v->root)) != NULL) { 958 node = rb_entry(rnode, struct xfrm_pol_inexact_node, node); 959 rb_erase(&node->node, &v->root); 960 xfrm_policy_inexact_node_reinsert(net, node, &n->root, 961 family); 962 } 963 964 hlist_for_each_entry(tmp, &v->hhead, bydst) { 965 tmp->bydst_reinsert = true; 966 hlist_del_rcu(&tmp->bydst); 967 } 968 969 xfrm_policy_inexact_list_reinsert(net, n, family); 970 } 971 972 static struct xfrm_pol_inexact_node * 973 xfrm_policy_inexact_insert_node(struct net *net, 974 struct rb_root *root, 975 xfrm_address_t *addr, 976 u16 family, u8 prefixlen, u8 dir) 977 { 978 struct xfrm_pol_inexact_node *cached = NULL; 979 struct rb_node **p, *parent = NULL; 980 struct xfrm_pol_inexact_node *node; 981 982 p = &root->rb_node; 983 while (*p) { 984 int delta; 985 986 parent = *p; 987 node = rb_entry(*p, struct xfrm_pol_inexact_node, node); 988 989 delta = xfrm_policy_addr_delta(addr, &node->addr, 990 node->prefixlen, 991 family); 992 if (delta == 0 && prefixlen >= node->prefixlen) { 993 WARN_ON_ONCE(cached); /* ipsec policies got lost */ 994 return node; 995 } 996 997 if (delta < 0) 998 p = &parent->rb_left; 999 else 1000 p = &parent->rb_right; 1001 1002 if (prefixlen < node->prefixlen) { 1003 delta = xfrm_policy_addr_delta(addr, &node->addr, 1004 prefixlen, 1005 family); 1006 if (delta) 1007 continue; 1008 1009 /* This node is a subnet of the new prefix. It needs 1010 * to be removed and re-inserted with the smaller 1011 * prefix and all nodes that are now also covered 1012 * by the reduced prefixlen. 1013 */ 1014 rb_erase(&node->node, root); 1015 1016 if (!cached) { 1017 xfrm_pol_inexact_node_init(node, addr, 1018 prefixlen); 1019 cached = node; 1020 } else { 1021 /* This node also falls within the new 1022 * prefixlen. Merge the to-be-reinserted 1023 * node and this one. 1024 */ 1025 xfrm_policy_inexact_node_merge(net, node, 1026 cached, family); 1027 kfree_rcu(node, rcu); 1028 } 1029 1030 /* restart */ 1031 p = &root->rb_node; 1032 parent = NULL; 1033 } 1034 } 1035 1036 node = cached; 1037 if (!node) { 1038 node = xfrm_pol_inexact_node_alloc(addr, prefixlen); 1039 if (!node) 1040 return NULL; 1041 } 1042 1043 rb_link_node_rcu(&node->node, parent, p); 1044 rb_insert_color(&node->node, root); 1045 1046 return node; 1047 } 1048 1049 static void xfrm_policy_inexact_gc_tree(struct rb_root *r, bool rm) 1050 { 1051 struct xfrm_pol_inexact_node *node; 1052 struct rb_node *rn = rb_first(r); 1053 1054 while (rn) { 1055 node = rb_entry(rn, struct xfrm_pol_inexact_node, node); 1056 1057 xfrm_policy_inexact_gc_tree(&node->root, rm); 1058 rn = rb_next(rn); 1059 1060 if (!hlist_empty(&node->hhead) || !RB_EMPTY_ROOT(&node->root)) { 1061 WARN_ON_ONCE(rm); 1062 continue; 1063 } 1064 1065 rb_erase(&node->node, r); 1066 kfree_rcu(node, rcu); 1067 } 1068 } 1069 1070 static void __xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b, bool net_exit) 1071 { 1072 write_seqcount_begin(&b->count); 1073 xfrm_policy_inexact_gc_tree(&b->root_d, net_exit); 1074 xfrm_policy_inexact_gc_tree(&b->root_s, net_exit); 1075 write_seqcount_end(&b->count); 1076 1077 if (!RB_EMPTY_ROOT(&b->root_d) || !RB_EMPTY_ROOT(&b->root_s) || 1078 !hlist_empty(&b->hhead)) { 1079 WARN_ON_ONCE(net_exit); 1080 return; 1081 } 1082 1083 if (rhashtable_remove_fast(&xfrm_policy_inexact_table, &b->head, 1084 xfrm_pol_inexact_params) == 0) { 1085 list_del(&b->inexact_bins); 1086 kfree_rcu(b, rcu); 1087 } 1088 } 1089 1090 static void xfrm_policy_inexact_prune_bin(struct xfrm_pol_inexact_bin *b) 1091 { 1092 struct net *net = read_pnet(&b->k.net); 1093 1094 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1095 __xfrm_policy_inexact_prune_bin(b, false); 1096 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1097 } 1098 1099 static void __xfrm_policy_inexact_flush(struct net *net) 1100 { 1101 struct xfrm_pol_inexact_bin *bin, *t; 1102 1103 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1104 1105 list_for_each_entry_safe(bin, t, &net->xfrm.inexact_bins, inexact_bins) 1106 __xfrm_policy_inexact_prune_bin(bin, false); 1107 } 1108 1109 static struct hlist_head * 1110 xfrm_policy_inexact_alloc_chain(struct xfrm_pol_inexact_bin *bin, 1111 struct xfrm_policy *policy, u8 dir) 1112 { 1113 struct xfrm_pol_inexact_node *n; 1114 struct net *net; 1115 1116 net = xp_net(policy); 1117 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1118 1119 if (xfrm_policy_inexact_insert_use_any_list(policy)) 1120 return &bin->hhead; 1121 1122 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.daddr, 1123 policy->family, 1124 policy->selector.prefixlen_d)) { 1125 write_seqcount_begin(&bin->count); 1126 n = xfrm_policy_inexact_insert_node(net, 1127 &bin->root_s, 1128 &policy->selector.saddr, 1129 policy->family, 1130 policy->selector.prefixlen_s, 1131 dir); 1132 write_seqcount_end(&bin->count); 1133 if (!n) 1134 return NULL; 1135 1136 return &n->hhead; 1137 } 1138 1139 /* daddr is fixed */ 1140 write_seqcount_begin(&bin->count); 1141 n = xfrm_policy_inexact_insert_node(net, 1142 &bin->root_d, 1143 &policy->selector.daddr, 1144 policy->family, 1145 policy->selector.prefixlen_d, dir); 1146 write_seqcount_end(&bin->count); 1147 if (!n) 1148 return NULL; 1149 1150 /* saddr is wildcard */ 1151 if (xfrm_pol_inexact_addr_use_any_list(&policy->selector.saddr, 1152 policy->family, 1153 policy->selector.prefixlen_s)) 1154 return &n->hhead; 1155 1156 write_seqcount_begin(&bin->count); 1157 n = xfrm_policy_inexact_insert_node(net, 1158 &n->root, 1159 &policy->selector.saddr, 1160 policy->family, 1161 policy->selector.prefixlen_s, dir); 1162 write_seqcount_end(&bin->count); 1163 if (!n) 1164 return NULL; 1165 1166 return &n->hhead; 1167 } 1168 1169 static struct xfrm_policy * 1170 xfrm_policy_inexact_insert(struct xfrm_policy *policy, u8 dir, int excl) 1171 { 1172 struct xfrm_pol_inexact_bin *bin; 1173 struct xfrm_policy *delpol; 1174 struct hlist_head *chain; 1175 struct net *net; 1176 1177 bin = xfrm_policy_inexact_alloc_bin(policy, dir); 1178 if (!bin) 1179 return ERR_PTR(-ENOMEM); 1180 1181 net = xp_net(policy); 1182 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 1183 1184 chain = xfrm_policy_inexact_alloc_chain(bin, policy, dir); 1185 if (!chain) { 1186 __xfrm_policy_inexact_prune_bin(bin, false); 1187 return ERR_PTR(-ENOMEM); 1188 } 1189 1190 delpol = xfrm_policy_insert_list(chain, policy, excl); 1191 if (delpol && excl) { 1192 __xfrm_policy_inexact_prune_bin(bin, false); 1193 return ERR_PTR(-EEXIST); 1194 } 1195 1196 chain = &net->xfrm.policy_inexact[dir]; 1197 xfrm_policy_insert_inexact_list(chain, policy); 1198 1199 if (delpol) 1200 __xfrm_policy_inexact_prune_bin(bin, false); 1201 1202 return delpol; 1203 } 1204 1205 static void xfrm_hash_rebuild(struct work_struct *work) 1206 { 1207 struct net *net = container_of(work, struct net, 1208 xfrm.policy_hthresh.work); 1209 unsigned int hmask; 1210 struct xfrm_policy *pol; 1211 struct xfrm_policy *policy; 1212 struct hlist_head *chain; 1213 struct hlist_head *odst; 1214 struct hlist_node *newpos; 1215 int i; 1216 int dir; 1217 unsigned seq; 1218 u8 lbits4, rbits4, lbits6, rbits6; 1219 1220 mutex_lock(&hash_resize_mutex); 1221 1222 /* read selector prefixlen thresholds */ 1223 do { 1224 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock); 1225 1226 lbits4 = net->xfrm.policy_hthresh.lbits4; 1227 rbits4 = net->xfrm.policy_hthresh.rbits4; 1228 lbits6 = net->xfrm.policy_hthresh.lbits6; 1229 rbits6 = net->xfrm.policy_hthresh.rbits6; 1230 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq)); 1231 1232 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1233 write_seqcount_begin(&xfrm_policy_hash_generation); 1234 1235 /* make sure that we can insert the indirect policies again before 1236 * we start with destructive action. 1237 */ 1238 list_for_each_entry(policy, &net->xfrm.policy_all, walk.all) { 1239 struct xfrm_pol_inexact_bin *bin; 1240 u8 dbits, sbits; 1241 1242 dir = xfrm_policy_id2dir(policy->index); 1243 if (policy->walk.dead || dir >= XFRM_POLICY_MAX) 1244 continue; 1245 1246 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) { 1247 if (policy->family == AF_INET) { 1248 dbits = rbits4; 1249 sbits = lbits4; 1250 } else { 1251 dbits = rbits6; 1252 sbits = lbits6; 1253 } 1254 } else { 1255 if (policy->family == AF_INET) { 1256 dbits = lbits4; 1257 sbits = rbits4; 1258 } else { 1259 dbits = lbits6; 1260 sbits = rbits6; 1261 } 1262 } 1263 1264 if (policy->selector.prefixlen_d < dbits || 1265 policy->selector.prefixlen_s < sbits) 1266 continue; 1267 1268 bin = xfrm_policy_inexact_alloc_bin(policy, dir); 1269 if (!bin) 1270 goto out_unlock; 1271 1272 if (!xfrm_policy_inexact_alloc_chain(bin, policy, dir)) 1273 goto out_unlock; 1274 } 1275 1276 /* reset the bydst and inexact table in all directions */ 1277 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 1278 struct hlist_node *n; 1279 1280 hlist_for_each_entry_safe(policy, n, 1281 &net->xfrm.policy_inexact[dir], 1282 bydst_inexact_list) { 1283 hlist_del_rcu(&policy->bydst); 1284 hlist_del_init(&policy->bydst_inexact_list); 1285 } 1286 1287 hmask = net->xfrm.policy_bydst[dir].hmask; 1288 odst = net->xfrm.policy_bydst[dir].table; 1289 for (i = hmask; i >= 0; i--) { 1290 hlist_for_each_entry_safe(policy, n, odst + i, bydst) 1291 hlist_del_rcu(&policy->bydst); 1292 } 1293 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) { 1294 /* dir out => dst = remote, src = local */ 1295 net->xfrm.policy_bydst[dir].dbits4 = rbits4; 1296 net->xfrm.policy_bydst[dir].sbits4 = lbits4; 1297 net->xfrm.policy_bydst[dir].dbits6 = rbits6; 1298 net->xfrm.policy_bydst[dir].sbits6 = lbits6; 1299 } else { 1300 /* dir in/fwd => dst = local, src = remote */ 1301 net->xfrm.policy_bydst[dir].dbits4 = lbits4; 1302 net->xfrm.policy_bydst[dir].sbits4 = rbits4; 1303 net->xfrm.policy_bydst[dir].dbits6 = lbits6; 1304 net->xfrm.policy_bydst[dir].sbits6 = rbits6; 1305 } 1306 } 1307 1308 /* re-insert all policies by order of creation */ 1309 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) { 1310 if (policy->walk.dead) 1311 continue; 1312 dir = xfrm_policy_id2dir(policy->index); 1313 if (dir >= XFRM_POLICY_MAX) { 1314 /* skip socket policies */ 1315 continue; 1316 } 1317 newpos = NULL; 1318 chain = policy_hash_bysel(net, &policy->selector, 1319 policy->family, dir); 1320 1321 if (!chain) { 1322 void *p = xfrm_policy_inexact_insert(policy, dir, 0); 1323 1324 WARN_ONCE(IS_ERR(p), "reinsert: %ld\n", PTR_ERR(p)); 1325 continue; 1326 } 1327 1328 hlist_for_each_entry(pol, chain, bydst) { 1329 if (policy->priority >= pol->priority) 1330 newpos = &pol->bydst; 1331 else 1332 break; 1333 } 1334 if (newpos) 1335 hlist_add_behind_rcu(&policy->bydst, newpos); 1336 else 1337 hlist_add_head_rcu(&policy->bydst, chain); 1338 } 1339 1340 out_unlock: 1341 __xfrm_policy_inexact_flush(net); 1342 write_seqcount_end(&xfrm_policy_hash_generation); 1343 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1344 1345 mutex_unlock(&hash_resize_mutex); 1346 } 1347 1348 void xfrm_policy_hash_rebuild(struct net *net) 1349 { 1350 schedule_work(&net->xfrm.policy_hthresh.work); 1351 } 1352 EXPORT_SYMBOL(xfrm_policy_hash_rebuild); 1353 1354 /* Generate new index... KAME seems to generate them ordered by cost 1355 * of an absolute inpredictability of ordering of rules. This will not pass. */ 1356 static u32 xfrm_gen_index(struct net *net, int dir, u32 index) 1357 { 1358 static u32 idx_generator; 1359 1360 for (;;) { 1361 struct hlist_head *list; 1362 struct xfrm_policy *p; 1363 u32 idx; 1364 int found; 1365 1366 if (!index) { 1367 idx = (idx_generator | dir); 1368 idx_generator += 8; 1369 } else { 1370 idx = index; 1371 index = 0; 1372 } 1373 1374 if (idx == 0) 1375 idx = 8; 1376 list = net->xfrm.policy_byidx + idx_hash(net, idx); 1377 found = 0; 1378 hlist_for_each_entry(p, list, byidx) { 1379 if (p->index == idx) { 1380 found = 1; 1381 break; 1382 } 1383 } 1384 if (!found) 1385 return idx; 1386 } 1387 } 1388 1389 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2) 1390 { 1391 u32 *p1 = (u32 *) s1; 1392 u32 *p2 = (u32 *) s2; 1393 int len = sizeof(struct xfrm_selector) / sizeof(u32); 1394 int i; 1395 1396 for (i = 0; i < len; i++) { 1397 if (p1[i] != p2[i]) 1398 return 1; 1399 } 1400 1401 return 0; 1402 } 1403 1404 static void xfrm_policy_requeue(struct xfrm_policy *old, 1405 struct xfrm_policy *new) 1406 { 1407 struct xfrm_policy_queue *pq = &old->polq; 1408 struct sk_buff_head list; 1409 1410 if (skb_queue_empty(&pq->hold_queue)) 1411 return; 1412 1413 __skb_queue_head_init(&list); 1414 1415 spin_lock_bh(&pq->hold_queue.lock); 1416 skb_queue_splice_init(&pq->hold_queue, &list); 1417 if (del_timer(&pq->hold_timer)) 1418 xfrm_pol_put(old); 1419 spin_unlock_bh(&pq->hold_queue.lock); 1420 1421 pq = &new->polq; 1422 1423 spin_lock_bh(&pq->hold_queue.lock); 1424 skb_queue_splice(&list, &pq->hold_queue); 1425 pq->timeout = XFRM_QUEUE_TMO_MIN; 1426 if (!mod_timer(&pq->hold_timer, jiffies)) 1427 xfrm_pol_hold(new); 1428 spin_unlock_bh(&pq->hold_queue.lock); 1429 } 1430 1431 static bool xfrm_policy_mark_match(struct xfrm_policy *policy, 1432 struct xfrm_policy *pol) 1433 { 1434 u32 mark = policy->mark.v & policy->mark.m; 1435 1436 if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m) 1437 return true; 1438 1439 if ((mark & pol->mark.m) == pol->mark.v && 1440 policy->priority == pol->priority) 1441 return true; 1442 1443 return false; 1444 } 1445 1446 static u32 xfrm_pol_bin_key(const void *data, u32 len, u32 seed) 1447 { 1448 const struct xfrm_pol_inexact_key *k = data; 1449 u32 a = k->type << 24 | k->dir << 16 | k->family; 1450 1451 return jhash_3words(a, k->if_id, net_hash_mix(read_pnet(&k->net)), 1452 seed); 1453 } 1454 1455 static u32 xfrm_pol_bin_obj(const void *data, u32 len, u32 seed) 1456 { 1457 const struct xfrm_pol_inexact_bin *b = data; 1458 1459 return xfrm_pol_bin_key(&b->k, 0, seed); 1460 } 1461 1462 static int xfrm_pol_bin_cmp(struct rhashtable_compare_arg *arg, 1463 const void *ptr) 1464 { 1465 const struct xfrm_pol_inexact_key *key = arg->key; 1466 const struct xfrm_pol_inexact_bin *b = ptr; 1467 int ret; 1468 1469 if (!net_eq(read_pnet(&b->k.net), read_pnet(&key->net))) 1470 return -1; 1471 1472 ret = b->k.dir ^ key->dir; 1473 if (ret) 1474 return ret; 1475 1476 ret = b->k.type ^ key->type; 1477 if (ret) 1478 return ret; 1479 1480 ret = b->k.family ^ key->family; 1481 if (ret) 1482 return ret; 1483 1484 return b->k.if_id ^ key->if_id; 1485 } 1486 1487 static const struct rhashtable_params xfrm_pol_inexact_params = { 1488 .head_offset = offsetof(struct xfrm_pol_inexact_bin, head), 1489 .hashfn = xfrm_pol_bin_key, 1490 .obj_hashfn = xfrm_pol_bin_obj, 1491 .obj_cmpfn = xfrm_pol_bin_cmp, 1492 .automatic_shrinking = true, 1493 }; 1494 1495 static void xfrm_policy_insert_inexact_list(struct hlist_head *chain, 1496 struct xfrm_policy *policy) 1497 { 1498 struct xfrm_policy *pol, *delpol = NULL; 1499 struct hlist_node *newpos = NULL; 1500 int i = 0; 1501 1502 hlist_for_each_entry(pol, chain, bydst_inexact_list) { 1503 if (pol->type == policy->type && 1504 pol->if_id == policy->if_id && 1505 !selector_cmp(&pol->selector, &policy->selector) && 1506 xfrm_policy_mark_match(policy, pol) && 1507 xfrm_sec_ctx_match(pol->security, policy->security) && 1508 !WARN_ON(delpol)) { 1509 delpol = pol; 1510 if (policy->priority > pol->priority) 1511 continue; 1512 } else if (policy->priority >= pol->priority) { 1513 newpos = &pol->bydst_inexact_list; 1514 continue; 1515 } 1516 if (delpol) 1517 break; 1518 } 1519 1520 if (newpos) 1521 hlist_add_behind_rcu(&policy->bydst_inexact_list, newpos); 1522 else 1523 hlist_add_head_rcu(&policy->bydst_inexact_list, chain); 1524 1525 hlist_for_each_entry(pol, chain, bydst_inexact_list) { 1526 pol->pos = i; 1527 i++; 1528 } 1529 } 1530 1531 static struct xfrm_policy *xfrm_policy_insert_list(struct hlist_head *chain, 1532 struct xfrm_policy *policy, 1533 bool excl) 1534 { 1535 struct xfrm_policy *pol, *newpos = NULL, *delpol = NULL; 1536 1537 hlist_for_each_entry(pol, chain, bydst) { 1538 if (pol->type == policy->type && 1539 pol->if_id == policy->if_id && 1540 !selector_cmp(&pol->selector, &policy->selector) && 1541 xfrm_policy_mark_match(policy, pol) && 1542 xfrm_sec_ctx_match(pol->security, policy->security) && 1543 !WARN_ON(delpol)) { 1544 if (excl) 1545 return ERR_PTR(-EEXIST); 1546 delpol = pol; 1547 if (policy->priority > pol->priority) 1548 continue; 1549 } else if (policy->priority >= pol->priority) { 1550 newpos = pol; 1551 continue; 1552 } 1553 if (delpol) 1554 break; 1555 } 1556 1557 if (newpos) 1558 hlist_add_behind_rcu(&policy->bydst, &newpos->bydst); 1559 else 1560 hlist_add_head_rcu(&policy->bydst, chain); 1561 1562 return delpol; 1563 } 1564 1565 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl) 1566 { 1567 struct net *net = xp_net(policy); 1568 struct xfrm_policy *delpol; 1569 struct hlist_head *chain; 1570 1571 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1572 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir); 1573 if (chain) 1574 delpol = xfrm_policy_insert_list(chain, policy, excl); 1575 else 1576 delpol = xfrm_policy_inexact_insert(policy, dir, excl); 1577 1578 if (IS_ERR(delpol)) { 1579 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1580 return PTR_ERR(delpol); 1581 } 1582 1583 __xfrm_policy_link(policy, dir); 1584 1585 /* After previous checking, family can either be AF_INET or AF_INET6 */ 1586 if (policy->family == AF_INET) 1587 rt_genid_bump_ipv4(net); 1588 else 1589 rt_genid_bump_ipv6(net); 1590 1591 if (delpol) { 1592 xfrm_policy_requeue(delpol, policy); 1593 __xfrm_policy_unlink(delpol, dir); 1594 } 1595 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index); 1596 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index)); 1597 policy->curlft.add_time = ktime_get_real_seconds(); 1598 policy->curlft.use_time = 0; 1599 if (!mod_timer(&policy->timer, jiffies + HZ)) 1600 xfrm_pol_hold(policy); 1601 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1602 1603 if (delpol) 1604 xfrm_policy_kill(delpol); 1605 else if (xfrm_bydst_should_resize(net, dir, NULL)) 1606 schedule_work(&net->xfrm.policy_hash_work); 1607 1608 return 0; 1609 } 1610 EXPORT_SYMBOL(xfrm_policy_insert); 1611 1612 static struct xfrm_policy * 1613 __xfrm_policy_bysel_ctx(struct hlist_head *chain, u32 mark, u32 if_id, 1614 u8 type, int dir, 1615 struct xfrm_selector *sel, 1616 struct xfrm_sec_ctx *ctx) 1617 { 1618 struct xfrm_policy *pol; 1619 1620 if (!chain) 1621 return NULL; 1622 1623 hlist_for_each_entry(pol, chain, bydst) { 1624 if (pol->type == type && 1625 pol->if_id == if_id && 1626 (mark & pol->mark.m) == pol->mark.v && 1627 !selector_cmp(sel, &pol->selector) && 1628 xfrm_sec_ctx_match(ctx, pol->security)) 1629 return pol; 1630 } 1631 1632 return NULL; 1633 } 1634 1635 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u32 if_id, 1636 u8 type, int dir, 1637 struct xfrm_selector *sel, 1638 struct xfrm_sec_ctx *ctx, int delete, 1639 int *err) 1640 { 1641 struct xfrm_pol_inexact_bin *bin = NULL; 1642 struct xfrm_policy *pol, *ret = NULL; 1643 struct hlist_head *chain; 1644 1645 *err = 0; 1646 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1647 chain = policy_hash_bysel(net, sel, sel->family, dir); 1648 if (!chain) { 1649 struct xfrm_pol_inexact_candidates cand; 1650 int i; 1651 1652 bin = xfrm_policy_inexact_lookup(net, type, 1653 sel->family, dir, if_id); 1654 if (!bin) { 1655 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1656 return NULL; 1657 } 1658 1659 if (!xfrm_policy_find_inexact_candidates(&cand, bin, 1660 &sel->saddr, 1661 &sel->daddr)) { 1662 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1663 return NULL; 1664 } 1665 1666 pol = NULL; 1667 for (i = 0; i < ARRAY_SIZE(cand.res); i++) { 1668 struct xfrm_policy *tmp; 1669 1670 tmp = __xfrm_policy_bysel_ctx(cand.res[i], mark, 1671 if_id, type, dir, 1672 sel, ctx); 1673 if (!tmp) 1674 continue; 1675 1676 if (!pol || tmp->pos < pol->pos) 1677 pol = tmp; 1678 } 1679 } else { 1680 pol = __xfrm_policy_bysel_ctx(chain, mark, if_id, type, dir, 1681 sel, ctx); 1682 } 1683 1684 if (pol) { 1685 xfrm_pol_hold(pol); 1686 if (delete) { 1687 *err = security_xfrm_policy_delete(pol->security); 1688 if (*err) { 1689 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1690 return pol; 1691 } 1692 __xfrm_policy_unlink(pol, dir); 1693 } 1694 ret = pol; 1695 } 1696 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1697 1698 if (ret && delete) 1699 xfrm_policy_kill(ret); 1700 if (bin && delete) 1701 xfrm_policy_inexact_prune_bin(bin); 1702 return ret; 1703 } 1704 EXPORT_SYMBOL(xfrm_policy_bysel_ctx); 1705 1706 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u32 if_id, 1707 u8 type, int dir, u32 id, int delete, 1708 int *err) 1709 { 1710 struct xfrm_policy *pol, *ret; 1711 struct hlist_head *chain; 1712 1713 *err = -ENOENT; 1714 if (xfrm_policy_id2dir(id) != dir) 1715 return NULL; 1716 1717 *err = 0; 1718 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1719 chain = net->xfrm.policy_byidx + idx_hash(net, id); 1720 ret = NULL; 1721 hlist_for_each_entry(pol, chain, byidx) { 1722 if (pol->type == type && pol->index == id && 1723 pol->if_id == if_id && 1724 (mark & pol->mark.m) == pol->mark.v) { 1725 xfrm_pol_hold(pol); 1726 if (delete) { 1727 *err = security_xfrm_policy_delete( 1728 pol->security); 1729 if (*err) { 1730 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1731 return pol; 1732 } 1733 __xfrm_policy_unlink(pol, dir); 1734 } 1735 ret = pol; 1736 break; 1737 } 1738 } 1739 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1740 1741 if (ret && delete) 1742 xfrm_policy_kill(ret); 1743 return ret; 1744 } 1745 EXPORT_SYMBOL(xfrm_policy_byid); 1746 1747 #ifdef CONFIG_SECURITY_NETWORK_XFRM 1748 static inline int 1749 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid) 1750 { 1751 struct xfrm_policy *pol; 1752 int err = 0; 1753 1754 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1755 if (pol->walk.dead || 1756 xfrm_policy_id2dir(pol->index) >= XFRM_POLICY_MAX || 1757 pol->type != type) 1758 continue; 1759 1760 err = security_xfrm_policy_delete(pol->security); 1761 if (err) { 1762 xfrm_audit_policy_delete(pol, 0, task_valid); 1763 return err; 1764 } 1765 } 1766 return err; 1767 } 1768 #else 1769 static inline int 1770 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid) 1771 { 1772 return 0; 1773 } 1774 #endif 1775 1776 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid) 1777 { 1778 int dir, err = 0, cnt = 0; 1779 struct xfrm_policy *pol; 1780 1781 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1782 1783 err = xfrm_policy_flush_secctx_check(net, type, task_valid); 1784 if (err) 1785 goto out; 1786 1787 again: 1788 list_for_each_entry(pol, &net->xfrm.policy_all, walk.all) { 1789 dir = xfrm_policy_id2dir(pol->index); 1790 if (pol->walk.dead || 1791 dir >= XFRM_POLICY_MAX || 1792 pol->type != type) 1793 continue; 1794 1795 __xfrm_policy_unlink(pol, dir); 1796 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1797 cnt++; 1798 xfrm_audit_policy_delete(pol, 1, task_valid); 1799 xfrm_policy_kill(pol); 1800 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1801 goto again; 1802 } 1803 if (cnt) 1804 __xfrm_policy_inexact_flush(net); 1805 else 1806 err = -ESRCH; 1807 out: 1808 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1809 return err; 1810 } 1811 EXPORT_SYMBOL(xfrm_policy_flush); 1812 1813 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk, 1814 int (*func)(struct xfrm_policy *, int, int, void*), 1815 void *data) 1816 { 1817 struct xfrm_policy *pol; 1818 struct xfrm_policy_walk_entry *x; 1819 int error = 0; 1820 1821 if (walk->type >= XFRM_POLICY_TYPE_MAX && 1822 walk->type != XFRM_POLICY_TYPE_ANY) 1823 return -EINVAL; 1824 1825 if (list_empty(&walk->walk.all) && walk->seq != 0) 1826 return 0; 1827 1828 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1829 if (list_empty(&walk->walk.all)) 1830 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all); 1831 else 1832 x = list_first_entry(&walk->walk.all, 1833 struct xfrm_policy_walk_entry, all); 1834 1835 list_for_each_entry_from(x, &net->xfrm.policy_all, all) { 1836 if (x->dead) 1837 continue; 1838 pol = container_of(x, struct xfrm_policy, walk); 1839 if (walk->type != XFRM_POLICY_TYPE_ANY && 1840 walk->type != pol->type) 1841 continue; 1842 error = func(pol, xfrm_policy_id2dir(pol->index), 1843 walk->seq, data); 1844 if (error) { 1845 list_move_tail(&walk->walk.all, &x->all); 1846 goto out; 1847 } 1848 walk->seq++; 1849 } 1850 if (walk->seq == 0) { 1851 error = -ENOENT; 1852 goto out; 1853 } 1854 list_del_init(&walk->walk.all); 1855 out: 1856 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1857 return error; 1858 } 1859 EXPORT_SYMBOL(xfrm_policy_walk); 1860 1861 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type) 1862 { 1863 INIT_LIST_HEAD(&walk->walk.all); 1864 walk->walk.dead = 1; 1865 walk->type = type; 1866 walk->seq = 0; 1867 } 1868 EXPORT_SYMBOL(xfrm_policy_walk_init); 1869 1870 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net) 1871 { 1872 if (list_empty(&walk->walk.all)) 1873 return; 1874 1875 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */ 1876 list_del(&walk->walk.all); 1877 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1878 } 1879 EXPORT_SYMBOL(xfrm_policy_walk_done); 1880 1881 /* 1882 * Find policy to apply to this flow. 1883 * 1884 * Returns 0 if policy found, else an -errno. 1885 */ 1886 static int xfrm_policy_match(const struct xfrm_policy *pol, 1887 const struct flowi *fl, 1888 u8 type, u16 family, int dir, u32 if_id) 1889 { 1890 const struct xfrm_selector *sel = &pol->selector; 1891 int ret = -ESRCH; 1892 bool match; 1893 1894 if (pol->family != family || 1895 pol->if_id != if_id || 1896 (fl->flowi_mark & pol->mark.m) != pol->mark.v || 1897 pol->type != type) 1898 return ret; 1899 1900 match = xfrm_selector_match(sel, fl, family); 1901 if (match) 1902 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid, 1903 dir); 1904 return ret; 1905 } 1906 1907 static struct xfrm_pol_inexact_node * 1908 xfrm_policy_lookup_inexact_addr(const struct rb_root *r, 1909 seqcount_t *count, 1910 const xfrm_address_t *addr, u16 family) 1911 { 1912 const struct rb_node *parent; 1913 int seq; 1914 1915 again: 1916 seq = read_seqcount_begin(count); 1917 1918 parent = rcu_dereference_raw(r->rb_node); 1919 while (parent) { 1920 struct xfrm_pol_inexact_node *node; 1921 int delta; 1922 1923 node = rb_entry(parent, struct xfrm_pol_inexact_node, node); 1924 1925 delta = xfrm_policy_addr_delta(addr, &node->addr, 1926 node->prefixlen, family); 1927 if (delta < 0) { 1928 parent = rcu_dereference_raw(parent->rb_left); 1929 continue; 1930 } else if (delta > 0) { 1931 parent = rcu_dereference_raw(parent->rb_right); 1932 continue; 1933 } 1934 1935 return node; 1936 } 1937 1938 if (read_seqcount_retry(count, seq)) 1939 goto again; 1940 1941 return NULL; 1942 } 1943 1944 static bool 1945 xfrm_policy_find_inexact_candidates(struct xfrm_pol_inexact_candidates *cand, 1946 struct xfrm_pol_inexact_bin *b, 1947 const xfrm_address_t *saddr, 1948 const xfrm_address_t *daddr) 1949 { 1950 struct xfrm_pol_inexact_node *n; 1951 u16 family; 1952 1953 if (!b) 1954 return false; 1955 1956 family = b->k.family; 1957 memset(cand, 0, sizeof(*cand)); 1958 cand->res[XFRM_POL_CAND_ANY] = &b->hhead; 1959 1960 n = xfrm_policy_lookup_inexact_addr(&b->root_d, &b->count, daddr, 1961 family); 1962 if (n) { 1963 cand->res[XFRM_POL_CAND_DADDR] = &n->hhead; 1964 n = xfrm_policy_lookup_inexact_addr(&n->root, &b->count, saddr, 1965 family); 1966 if (n) 1967 cand->res[XFRM_POL_CAND_BOTH] = &n->hhead; 1968 } 1969 1970 n = xfrm_policy_lookup_inexact_addr(&b->root_s, &b->count, saddr, 1971 family); 1972 if (n) 1973 cand->res[XFRM_POL_CAND_SADDR] = &n->hhead; 1974 1975 return true; 1976 } 1977 1978 static struct xfrm_pol_inexact_bin * 1979 xfrm_policy_inexact_lookup_rcu(struct net *net, u8 type, u16 family, 1980 u8 dir, u32 if_id) 1981 { 1982 struct xfrm_pol_inexact_key k = { 1983 .family = family, 1984 .type = type, 1985 .dir = dir, 1986 .if_id = if_id, 1987 }; 1988 1989 write_pnet(&k.net, net); 1990 1991 return rhashtable_lookup(&xfrm_policy_inexact_table, &k, 1992 xfrm_pol_inexact_params); 1993 } 1994 1995 static struct xfrm_pol_inexact_bin * 1996 xfrm_policy_inexact_lookup(struct net *net, u8 type, u16 family, 1997 u8 dir, u32 if_id) 1998 { 1999 struct xfrm_pol_inexact_bin *bin; 2000 2001 lockdep_assert_held(&net->xfrm.xfrm_policy_lock); 2002 2003 rcu_read_lock(); 2004 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id); 2005 rcu_read_unlock(); 2006 2007 return bin; 2008 } 2009 2010 static struct xfrm_policy * 2011 __xfrm_policy_eval_candidates(struct hlist_head *chain, 2012 struct xfrm_policy *prefer, 2013 const struct flowi *fl, 2014 u8 type, u16 family, int dir, u32 if_id) 2015 { 2016 u32 priority = prefer ? prefer->priority : ~0u; 2017 struct xfrm_policy *pol; 2018 2019 if (!chain) 2020 return NULL; 2021 2022 hlist_for_each_entry_rcu(pol, chain, bydst) { 2023 int err; 2024 2025 if (pol->priority > priority) 2026 break; 2027 2028 err = xfrm_policy_match(pol, fl, type, family, dir, if_id); 2029 if (err) { 2030 if (err != -ESRCH) 2031 return ERR_PTR(err); 2032 2033 continue; 2034 } 2035 2036 if (prefer) { 2037 /* matches. Is it older than *prefer? */ 2038 if (pol->priority == priority && 2039 prefer->pos < pol->pos) 2040 return prefer; 2041 } 2042 2043 return pol; 2044 } 2045 2046 return NULL; 2047 } 2048 2049 static struct xfrm_policy * 2050 xfrm_policy_eval_candidates(struct xfrm_pol_inexact_candidates *cand, 2051 struct xfrm_policy *prefer, 2052 const struct flowi *fl, 2053 u8 type, u16 family, int dir, u32 if_id) 2054 { 2055 struct xfrm_policy *tmp; 2056 int i; 2057 2058 for (i = 0; i < ARRAY_SIZE(cand->res); i++) { 2059 tmp = __xfrm_policy_eval_candidates(cand->res[i], 2060 prefer, 2061 fl, type, family, dir, 2062 if_id); 2063 if (!tmp) 2064 continue; 2065 2066 if (IS_ERR(tmp)) 2067 return tmp; 2068 prefer = tmp; 2069 } 2070 2071 return prefer; 2072 } 2073 2074 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type, 2075 const struct flowi *fl, 2076 u16 family, u8 dir, 2077 u32 if_id) 2078 { 2079 struct xfrm_pol_inexact_candidates cand; 2080 const xfrm_address_t *daddr, *saddr; 2081 struct xfrm_pol_inexact_bin *bin; 2082 struct xfrm_policy *pol, *ret; 2083 struct hlist_head *chain; 2084 unsigned int sequence; 2085 int err; 2086 2087 daddr = xfrm_flowi_daddr(fl, family); 2088 saddr = xfrm_flowi_saddr(fl, family); 2089 if (unlikely(!daddr || !saddr)) 2090 return NULL; 2091 2092 rcu_read_lock(); 2093 retry: 2094 do { 2095 sequence = read_seqcount_begin(&xfrm_policy_hash_generation); 2096 chain = policy_hash_direct(net, daddr, saddr, family, dir); 2097 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence)); 2098 2099 ret = NULL; 2100 hlist_for_each_entry_rcu(pol, chain, bydst) { 2101 err = xfrm_policy_match(pol, fl, type, family, dir, if_id); 2102 if (err) { 2103 if (err == -ESRCH) 2104 continue; 2105 else { 2106 ret = ERR_PTR(err); 2107 goto fail; 2108 } 2109 } else { 2110 ret = pol; 2111 break; 2112 } 2113 } 2114 bin = xfrm_policy_inexact_lookup_rcu(net, type, family, dir, if_id); 2115 if (!bin || !xfrm_policy_find_inexact_candidates(&cand, bin, saddr, 2116 daddr)) 2117 goto skip_inexact; 2118 2119 pol = xfrm_policy_eval_candidates(&cand, ret, fl, type, 2120 family, dir, if_id); 2121 if (pol) { 2122 ret = pol; 2123 if (IS_ERR(pol)) 2124 goto fail; 2125 } 2126 2127 skip_inexact: 2128 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence)) 2129 goto retry; 2130 2131 if (ret && !xfrm_pol_hold_rcu(ret)) 2132 goto retry; 2133 fail: 2134 rcu_read_unlock(); 2135 2136 return ret; 2137 } 2138 2139 static struct xfrm_policy *xfrm_policy_lookup(struct net *net, 2140 const struct flowi *fl, 2141 u16 family, u8 dir, u32 if_id) 2142 { 2143 #ifdef CONFIG_XFRM_SUB_POLICY 2144 struct xfrm_policy *pol; 2145 2146 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, 2147 dir, if_id); 2148 if (pol != NULL) 2149 return pol; 2150 #endif 2151 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, 2152 dir, if_id); 2153 } 2154 2155 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir, 2156 const struct flowi *fl, 2157 u16 family, u32 if_id) 2158 { 2159 struct xfrm_policy *pol; 2160 2161 rcu_read_lock(); 2162 again: 2163 pol = rcu_dereference(sk->sk_policy[dir]); 2164 if (pol != NULL) { 2165 bool match; 2166 int err = 0; 2167 2168 if (pol->family != family) { 2169 pol = NULL; 2170 goto out; 2171 } 2172 2173 match = xfrm_selector_match(&pol->selector, fl, family); 2174 if (match) { 2175 if ((sk->sk_mark & pol->mark.m) != pol->mark.v || 2176 pol->if_id != if_id) { 2177 pol = NULL; 2178 goto out; 2179 } 2180 err = security_xfrm_policy_lookup(pol->security, 2181 fl->flowi_secid, 2182 dir); 2183 if (!err) { 2184 if (!xfrm_pol_hold_rcu(pol)) 2185 goto again; 2186 } else if (err == -ESRCH) { 2187 pol = NULL; 2188 } else { 2189 pol = ERR_PTR(err); 2190 } 2191 } else 2192 pol = NULL; 2193 } 2194 out: 2195 rcu_read_unlock(); 2196 return pol; 2197 } 2198 2199 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir) 2200 { 2201 struct net *net = xp_net(pol); 2202 2203 list_add(&pol->walk.all, &net->xfrm.policy_all); 2204 net->xfrm.policy_count[dir]++; 2205 xfrm_pol_hold(pol); 2206 } 2207 2208 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol, 2209 int dir) 2210 { 2211 struct net *net = xp_net(pol); 2212 2213 if (list_empty(&pol->walk.all)) 2214 return NULL; 2215 2216 /* Socket policies are not hashed. */ 2217 if (!hlist_unhashed(&pol->bydst)) { 2218 hlist_del_rcu(&pol->bydst); 2219 hlist_del_init(&pol->bydst_inexact_list); 2220 hlist_del(&pol->byidx); 2221 } 2222 2223 list_del_init(&pol->walk.all); 2224 net->xfrm.policy_count[dir]--; 2225 2226 return pol; 2227 } 2228 2229 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir) 2230 { 2231 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir); 2232 } 2233 2234 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir) 2235 { 2236 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir); 2237 } 2238 2239 int xfrm_policy_delete(struct xfrm_policy *pol, int dir) 2240 { 2241 struct net *net = xp_net(pol); 2242 2243 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 2244 pol = __xfrm_policy_unlink(pol, dir); 2245 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 2246 if (pol) { 2247 xfrm_policy_kill(pol); 2248 return 0; 2249 } 2250 return -ENOENT; 2251 } 2252 EXPORT_SYMBOL(xfrm_policy_delete); 2253 2254 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol) 2255 { 2256 struct net *net = sock_net(sk); 2257 struct xfrm_policy *old_pol; 2258 2259 #ifdef CONFIG_XFRM_SUB_POLICY 2260 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN) 2261 return -EINVAL; 2262 #endif 2263 2264 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 2265 old_pol = rcu_dereference_protected(sk->sk_policy[dir], 2266 lockdep_is_held(&net->xfrm.xfrm_policy_lock)); 2267 if (pol) { 2268 pol->curlft.add_time = ktime_get_real_seconds(); 2269 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0); 2270 xfrm_sk_policy_link(pol, dir); 2271 } 2272 rcu_assign_pointer(sk->sk_policy[dir], pol); 2273 if (old_pol) { 2274 if (pol) 2275 xfrm_policy_requeue(old_pol, pol); 2276 2277 /* Unlinking succeeds always. This is the only function 2278 * allowed to delete or replace socket policy. 2279 */ 2280 xfrm_sk_policy_unlink(old_pol, dir); 2281 } 2282 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 2283 2284 if (old_pol) { 2285 xfrm_policy_kill(old_pol); 2286 } 2287 return 0; 2288 } 2289 2290 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir) 2291 { 2292 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC); 2293 struct net *net = xp_net(old); 2294 2295 if (newp) { 2296 newp->selector = old->selector; 2297 if (security_xfrm_policy_clone(old->security, 2298 &newp->security)) { 2299 kfree(newp); 2300 return NULL; /* ENOMEM */ 2301 } 2302 newp->lft = old->lft; 2303 newp->curlft = old->curlft; 2304 newp->mark = old->mark; 2305 newp->if_id = old->if_id; 2306 newp->action = old->action; 2307 newp->flags = old->flags; 2308 newp->xfrm_nr = old->xfrm_nr; 2309 newp->index = old->index; 2310 newp->type = old->type; 2311 newp->family = old->family; 2312 memcpy(newp->xfrm_vec, old->xfrm_vec, 2313 newp->xfrm_nr*sizeof(struct xfrm_tmpl)); 2314 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 2315 xfrm_sk_policy_link(newp, dir); 2316 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 2317 xfrm_pol_put(newp); 2318 } 2319 return newp; 2320 } 2321 2322 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk) 2323 { 2324 const struct xfrm_policy *p; 2325 struct xfrm_policy *np; 2326 int i, ret = 0; 2327 2328 rcu_read_lock(); 2329 for (i = 0; i < 2; i++) { 2330 p = rcu_dereference(osk->sk_policy[i]); 2331 if (p) { 2332 np = clone_policy(p, i); 2333 if (unlikely(!np)) { 2334 ret = -ENOMEM; 2335 break; 2336 } 2337 rcu_assign_pointer(sk->sk_policy[i], np); 2338 } 2339 } 2340 rcu_read_unlock(); 2341 return ret; 2342 } 2343 2344 static int 2345 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local, 2346 xfrm_address_t *remote, unsigned short family, u32 mark) 2347 { 2348 int err; 2349 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 2350 2351 if (unlikely(afinfo == NULL)) 2352 return -EINVAL; 2353 err = afinfo->get_saddr(net, oif, local, remote, mark); 2354 rcu_read_unlock(); 2355 return err; 2356 } 2357 2358 /* Resolve list of templates for the flow, given policy. */ 2359 2360 static int 2361 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl, 2362 struct xfrm_state **xfrm, unsigned short family) 2363 { 2364 struct net *net = xp_net(policy); 2365 int nx; 2366 int i, error; 2367 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family); 2368 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family); 2369 xfrm_address_t tmp; 2370 2371 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) { 2372 struct xfrm_state *x; 2373 xfrm_address_t *remote = daddr; 2374 xfrm_address_t *local = saddr; 2375 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i]; 2376 2377 if (tmpl->mode == XFRM_MODE_TUNNEL || 2378 tmpl->mode == XFRM_MODE_BEET) { 2379 remote = &tmpl->id.daddr; 2380 local = &tmpl->saddr; 2381 if (xfrm_addr_any(local, tmpl->encap_family)) { 2382 error = xfrm_get_saddr(net, fl->flowi_oif, 2383 &tmp, remote, 2384 tmpl->encap_family, 0); 2385 if (error) 2386 goto fail; 2387 local = &tmp; 2388 } 2389 } 2390 2391 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, 2392 family, policy->if_id); 2393 2394 if (x && x->km.state == XFRM_STATE_VALID) { 2395 xfrm[nx++] = x; 2396 daddr = remote; 2397 saddr = local; 2398 continue; 2399 } 2400 if (x) { 2401 error = (x->km.state == XFRM_STATE_ERROR ? 2402 -EINVAL : -EAGAIN); 2403 xfrm_state_put(x); 2404 } else if (error == -ESRCH) { 2405 error = -EAGAIN; 2406 } 2407 2408 if (!tmpl->optional) 2409 goto fail; 2410 } 2411 return nx; 2412 2413 fail: 2414 for (nx--; nx >= 0; nx--) 2415 xfrm_state_put(xfrm[nx]); 2416 return error; 2417 } 2418 2419 static int 2420 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl, 2421 struct xfrm_state **xfrm, unsigned short family) 2422 { 2423 struct xfrm_state *tp[XFRM_MAX_DEPTH]; 2424 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm; 2425 int cnx = 0; 2426 int error; 2427 int ret; 2428 int i; 2429 2430 for (i = 0; i < npols; i++) { 2431 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) { 2432 error = -ENOBUFS; 2433 goto fail; 2434 } 2435 2436 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family); 2437 if (ret < 0) { 2438 error = ret; 2439 goto fail; 2440 } else 2441 cnx += ret; 2442 } 2443 2444 /* found states are sorted for outbound processing */ 2445 if (npols > 1) 2446 xfrm_state_sort(xfrm, tpp, cnx, family); 2447 2448 return cnx; 2449 2450 fail: 2451 for (cnx--; cnx >= 0; cnx--) 2452 xfrm_state_put(tpp[cnx]); 2453 return error; 2454 2455 } 2456 2457 static int xfrm_get_tos(const struct flowi *fl, int family) 2458 { 2459 if (family == AF_INET) 2460 return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos; 2461 2462 return 0; 2463 } 2464 2465 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family) 2466 { 2467 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 2468 struct dst_ops *dst_ops; 2469 struct xfrm_dst *xdst; 2470 2471 if (!afinfo) 2472 return ERR_PTR(-EINVAL); 2473 2474 switch (family) { 2475 case AF_INET: 2476 dst_ops = &net->xfrm.xfrm4_dst_ops; 2477 break; 2478 #if IS_ENABLED(CONFIG_IPV6) 2479 case AF_INET6: 2480 dst_ops = &net->xfrm.xfrm6_dst_ops; 2481 break; 2482 #endif 2483 default: 2484 BUG(); 2485 } 2486 xdst = dst_alloc(dst_ops, NULL, 1, DST_OBSOLETE_NONE, 0); 2487 2488 if (likely(xdst)) { 2489 struct dst_entry *dst = &xdst->u.dst; 2490 2491 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*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 2597 if (xfrm[i]->props.smark.v || xfrm[i]->props.smark.m) 2598 mark = xfrm_smark_get(fl->flowi_mark, xfrm[i]); 2599 2600 family = xfrm[i]->props.family; 2601 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif, 2602 &saddr, &daddr, family, mark); 2603 err = PTR_ERR(dst); 2604 if (IS_ERR(dst)) 2605 goto put_states; 2606 } else 2607 dst_hold(dst); 2608 2609 dst1->xfrm = xfrm[i]; 2610 xdst->xfrm_genid = xfrm[i]->genid; 2611 2612 dst1->obsolete = DST_OBSOLETE_FORCE_CHK; 2613 dst1->flags |= DST_HOST; 2614 dst1->lastuse = now; 2615 2616 dst1->input = dst_discard; 2617 2618 rcu_read_lock(); 2619 afinfo = xfrm_state_afinfo_get_rcu(inner_mode->family); 2620 if (likely(afinfo)) 2621 dst1->output = afinfo->output; 2622 else 2623 dst1->output = dst_discard_out; 2624 rcu_read_unlock(); 2625 2626 xdst_prev = xdst; 2627 2628 header_len += xfrm[i]->props.header_len; 2629 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT) 2630 nfheader_len += xfrm[i]->props.header_len; 2631 trailer_len += xfrm[i]->props.trailer_len; 2632 } 2633 2634 xfrm_dst_set_child(xdst_prev, dst); 2635 xdst0->path = dst; 2636 2637 err = -ENODEV; 2638 dev = dst->dev; 2639 if (!dev) 2640 goto free_dst; 2641 2642 xfrm_init_path(xdst0, dst, nfheader_len); 2643 xfrm_init_pmtu(bundle, nx); 2644 2645 for (xdst_prev = xdst0; xdst_prev != (struct xfrm_dst *)dst; 2646 xdst_prev = (struct xfrm_dst *) xfrm_dst_child(&xdst_prev->u.dst)) { 2647 err = xfrm_fill_dst(xdst_prev, dev, fl); 2648 if (err) 2649 goto free_dst; 2650 2651 xdst_prev->u.dst.header_len = header_len; 2652 xdst_prev->u.dst.trailer_len = trailer_len; 2653 header_len -= xdst_prev->u.dst.xfrm->props.header_len; 2654 trailer_len -= xdst_prev->u.dst.xfrm->props.trailer_len; 2655 } 2656 2657 return &xdst0->u.dst; 2658 2659 put_states: 2660 for (; i < nx; i++) 2661 xfrm_state_put(xfrm[i]); 2662 free_dst: 2663 if (xdst0) 2664 dst_release_immediate(&xdst0->u.dst); 2665 2666 return ERR_PTR(err); 2667 } 2668 2669 static int xfrm_expand_policies(const struct flowi *fl, u16 family, 2670 struct xfrm_policy **pols, 2671 int *num_pols, int *num_xfrms) 2672 { 2673 int i; 2674 2675 if (*num_pols == 0 || !pols[0]) { 2676 *num_pols = 0; 2677 *num_xfrms = 0; 2678 return 0; 2679 } 2680 if (IS_ERR(pols[0])) 2681 return PTR_ERR(pols[0]); 2682 2683 *num_xfrms = pols[0]->xfrm_nr; 2684 2685 #ifdef CONFIG_XFRM_SUB_POLICY 2686 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW && 2687 pols[0]->type != XFRM_POLICY_TYPE_MAIN) { 2688 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]), 2689 XFRM_POLICY_TYPE_MAIN, 2690 fl, family, 2691 XFRM_POLICY_OUT, 2692 pols[0]->if_id); 2693 if (pols[1]) { 2694 if (IS_ERR(pols[1])) { 2695 xfrm_pols_put(pols, *num_pols); 2696 return PTR_ERR(pols[1]); 2697 } 2698 (*num_pols)++; 2699 (*num_xfrms) += pols[1]->xfrm_nr; 2700 } 2701 } 2702 #endif 2703 for (i = 0; i < *num_pols; i++) { 2704 if (pols[i]->action != XFRM_POLICY_ALLOW) { 2705 *num_xfrms = -1; 2706 break; 2707 } 2708 } 2709 2710 return 0; 2711 2712 } 2713 2714 static struct xfrm_dst * 2715 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols, 2716 const struct flowi *fl, u16 family, 2717 struct dst_entry *dst_orig) 2718 { 2719 struct net *net = xp_net(pols[0]); 2720 struct xfrm_state *xfrm[XFRM_MAX_DEPTH]; 2721 struct xfrm_dst *bundle[XFRM_MAX_DEPTH]; 2722 struct xfrm_dst *xdst; 2723 struct dst_entry *dst; 2724 int err; 2725 2726 /* Try to instantiate a bundle */ 2727 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family); 2728 if (err <= 0) { 2729 if (err == 0) 2730 return NULL; 2731 2732 if (err != -EAGAIN) 2733 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR); 2734 return ERR_PTR(err); 2735 } 2736 2737 dst = xfrm_bundle_create(pols[0], xfrm, bundle, err, fl, dst_orig); 2738 if (IS_ERR(dst)) { 2739 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR); 2740 return ERR_CAST(dst); 2741 } 2742 2743 xdst = (struct xfrm_dst *)dst; 2744 xdst->num_xfrms = err; 2745 xdst->num_pols = num_pols; 2746 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols); 2747 xdst->policy_genid = atomic_read(&pols[0]->genid); 2748 2749 return xdst; 2750 } 2751 2752 static void xfrm_policy_queue_process(struct timer_list *t) 2753 { 2754 struct sk_buff *skb; 2755 struct sock *sk; 2756 struct dst_entry *dst; 2757 struct xfrm_policy *pol = from_timer(pol, t, polq.hold_timer); 2758 struct net *net = xp_net(pol); 2759 struct xfrm_policy_queue *pq = &pol->polq; 2760 struct flowi fl; 2761 struct sk_buff_head list; 2762 2763 spin_lock(&pq->hold_queue.lock); 2764 skb = skb_peek(&pq->hold_queue); 2765 if (!skb) { 2766 spin_unlock(&pq->hold_queue.lock); 2767 goto out; 2768 } 2769 dst = skb_dst(skb); 2770 sk = skb->sk; 2771 xfrm_decode_session(skb, &fl, dst->ops->family); 2772 spin_unlock(&pq->hold_queue.lock); 2773 2774 dst_hold(xfrm_dst_path(dst)); 2775 dst = xfrm_lookup(net, xfrm_dst_path(dst), &fl, sk, XFRM_LOOKUP_QUEUE); 2776 if (IS_ERR(dst)) 2777 goto purge_queue; 2778 2779 if (dst->flags & DST_XFRM_QUEUE) { 2780 dst_release(dst); 2781 2782 if (pq->timeout >= XFRM_QUEUE_TMO_MAX) 2783 goto purge_queue; 2784 2785 pq->timeout = pq->timeout << 1; 2786 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout)) 2787 xfrm_pol_hold(pol); 2788 goto out; 2789 } 2790 2791 dst_release(dst); 2792 2793 __skb_queue_head_init(&list); 2794 2795 spin_lock(&pq->hold_queue.lock); 2796 pq->timeout = 0; 2797 skb_queue_splice_init(&pq->hold_queue, &list); 2798 spin_unlock(&pq->hold_queue.lock); 2799 2800 while (!skb_queue_empty(&list)) { 2801 skb = __skb_dequeue(&list); 2802 2803 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family); 2804 dst_hold(xfrm_dst_path(skb_dst(skb))); 2805 dst = xfrm_lookup(net, xfrm_dst_path(skb_dst(skb)), &fl, skb->sk, 0); 2806 if (IS_ERR(dst)) { 2807 kfree_skb(skb); 2808 continue; 2809 } 2810 2811 nf_reset_ct(skb); 2812 skb_dst_drop(skb); 2813 skb_dst_set(skb, dst); 2814 2815 dst_output(net, skb->sk, skb); 2816 } 2817 2818 out: 2819 xfrm_pol_put(pol); 2820 return; 2821 2822 purge_queue: 2823 pq->timeout = 0; 2824 skb_queue_purge(&pq->hold_queue); 2825 xfrm_pol_put(pol); 2826 } 2827 2828 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb) 2829 { 2830 unsigned long sched_next; 2831 struct dst_entry *dst = skb_dst(skb); 2832 struct xfrm_dst *xdst = (struct xfrm_dst *) dst; 2833 struct xfrm_policy *pol = xdst->pols[0]; 2834 struct xfrm_policy_queue *pq = &pol->polq; 2835 2836 if (unlikely(skb_fclone_busy(sk, skb))) { 2837 kfree_skb(skb); 2838 return 0; 2839 } 2840 2841 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) { 2842 kfree_skb(skb); 2843 return -EAGAIN; 2844 } 2845 2846 skb_dst_force(skb); 2847 2848 spin_lock_bh(&pq->hold_queue.lock); 2849 2850 if (!pq->timeout) 2851 pq->timeout = XFRM_QUEUE_TMO_MIN; 2852 2853 sched_next = jiffies + pq->timeout; 2854 2855 if (del_timer(&pq->hold_timer)) { 2856 if (time_before(pq->hold_timer.expires, sched_next)) 2857 sched_next = pq->hold_timer.expires; 2858 xfrm_pol_put(pol); 2859 } 2860 2861 __skb_queue_tail(&pq->hold_queue, skb); 2862 if (!mod_timer(&pq->hold_timer, sched_next)) 2863 xfrm_pol_hold(pol); 2864 2865 spin_unlock_bh(&pq->hold_queue.lock); 2866 2867 return 0; 2868 } 2869 2870 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net, 2871 struct xfrm_flo *xflo, 2872 const struct flowi *fl, 2873 int num_xfrms, 2874 u16 family) 2875 { 2876 int err; 2877 struct net_device *dev; 2878 struct dst_entry *dst; 2879 struct dst_entry *dst1; 2880 struct xfrm_dst *xdst; 2881 2882 xdst = xfrm_alloc_dst(net, family); 2883 if (IS_ERR(xdst)) 2884 return xdst; 2885 2886 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) || 2887 net->xfrm.sysctl_larval_drop || 2888 num_xfrms <= 0) 2889 return xdst; 2890 2891 dst = xflo->dst_orig; 2892 dst1 = &xdst->u.dst; 2893 dst_hold(dst); 2894 xdst->route = dst; 2895 2896 dst_copy_metrics(dst1, dst); 2897 2898 dst1->obsolete = DST_OBSOLETE_FORCE_CHK; 2899 dst1->flags |= DST_HOST | DST_XFRM_QUEUE; 2900 dst1->lastuse = jiffies; 2901 2902 dst1->input = dst_discard; 2903 dst1->output = xdst_queue_output; 2904 2905 dst_hold(dst); 2906 xfrm_dst_set_child(xdst, dst); 2907 xdst->path = dst; 2908 2909 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0); 2910 2911 err = -ENODEV; 2912 dev = dst->dev; 2913 if (!dev) 2914 goto free_dst; 2915 2916 err = xfrm_fill_dst(xdst, dev, fl); 2917 if (err) 2918 goto free_dst; 2919 2920 out: 2921 return xdst; 2922 2923 free_dst: 2924 dst_release(dst1); 2925 xdst = ERR_PTR(err); 2926 goto out; 2927 } 2928 2929 static struct xfrm_dst *xfrm_bundle_lookup(struct net *net, 2930 const struct flowi *fl, 2931 u16 family, u8 dir, 2932 struct xfrm_flo *xflo, u32 if_id) 2933 { 2934 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 2935 int num_pols = 0, num_xfrms = 0, err; 2936 struct xfrm_dst *xdst; 2937 2938 /* Resolve policies to use if we couldn't get them from 2939 * previous cache entry */ 2940 num_pols = 1; 2941 pols[0] = xfrm_policy_lookup(net, fl, family, dir, if_id); 2942 err = xfrm_expand_policies(fl, family, pols, 2943 &num_pols, &num_xfrms); 2944 if (err < 0) 2945 goto inc_error; 2946 if (num_pols == 0) 2947 return NULL; 2948 if (num_xfrms <= 0) 2949 goto make_dummy_bundle; 2950 2951 xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, 2952 xflo->dst_orig); 2953 if (IS_ERR(xdst)) { 2954 err = PTR_ERR(xdst); 2955 if (err == -EREMOTE) { 2956 xfrm_pols_put(pols, num_pols); 2957 return NULL; 2958 } 2959 2960 if (err != -EAGAIN) 2961 goto error; 2962 goto make_dummy_bundle; 2963 } else if (xdst == NULL) { 2964 num_xfrms = 0; 2965 goto make_dummy_bundle; 2966 } 2967 2968 return xdst; 2969 2970 make_dummy_bundle: 2971 /* We found policies, but there's no bundles to instantiate: 2972 * either because the policy blocks, has no transformations or 2973 * we could not build template (no xfrm_states).*/ 2974 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family); 2975 if (IS_ERR(xdst)) { 2976 xfrm_pols_put(pols, num_pols); 2977 return ERR_CAST(xdst); 2978 } 2979 xdst->num_pols = num_pols; 2980 xdst->num_xfrms = num_xfrms; 2981 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols); 2982 2983 return xdst; 2984 2985 inc_error: 2986 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR); 2987 error: 2988 xfrm_pols_put(pols, num_pols); 2989 return ERR_PTR(err); 2990 } 2991 2992 static struct dst_entry *make_blackhole(struct net *net, u16 family, 2993 struct dst_entry *dst_orig) 2994 { 2995 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 2996 struct dst_entry *ret; 2997 2998 if (!afinfo) { 2999 dst_release(dst_orig); 3000 return ERR_PTR(-EINVAL); 3001 } else { 3002 ret = afinfo->blackhole_route(net, dst_orig); 3003 } 3004 rcu_read_unlock(); 3005 3006 return ret; 3007 } 3008 3009 /* Finds/creates a bundle for given flow and if_id 3010 * 3011 * At the moment we eat a raw IP route. Mostly to speed up lookups 3012 * on interfaces with disabled IPsec. 3013 * 3014 * xfrm_lookup uses an if_id of 0 by default, and is provided for 3015 * compatibility 3016 */ 3017 struct dst_entry *xfrm_lookup_with_ifid(struct net *net, 3018 struct dst_entry *dst_orig, 3019 const struct flowi *fl, 3020 const struct sock *sk, 3021 int flags, u32 if_id) 3022 { 3023 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 3024 struct xfrm_dst *xdst; 3025 struct dst_entry *dst, *route; 3026 u16 family = dst_orig->ops->family; 3027 u8 dir = XFRM_POLICY_OUT; 3028 int i, err, num_pols, num_xfrms = 0, drop_pols = 0; 3029 3030 dst = NULL; 3031 xdst = NULL; 3032 route = NULL; 3033 3034 sk = sk_const_to_full_sk(sk); 3035 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) { 3036 num_pols = 1; 3037 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family, 3038 if_id); 3039 err = xfrm_expand_policies(fl, family, pols, 3040 &num_pols, &num_xfrms); 3041 if (err < 0) 3042 goto dropdst; 3043 3044 if (num_pols) { 3045 if (num_xfrms <= 0) { 3046 drop_pols = num_pols; 3047 goto no_transform; 3048 } 3049 3050 xdst = xfrm_resolve_and_create_bundle( 3051 pols, num_pols, fl, 3052 family, dst_orig); 3053 3054 if (IS_ERR(xdst)) { 3055 xfrm_pols_put(pols, num_pols); 3056 err = PTR_ERR(xdst); 3057 if (err == -EREMOTE) 3058 goto nopol; 3059 3060 goto dropdst; 3061 } else if (xdst == NULL) { 3062 num_xfrms = 0; 3063 drop_pols = num_pols; 3064 goto no_transform; 3065 } 3066 3067 route = xdst->route; 3068 } 3069 } 3070 3071 if (xdst == NULL) { 3072 struct xfrm_flo xflo; 3073 3074 xflo.dst_orig = dst_orig; 3075 xflo.flags = flags; 3076 3077 /* To accelerate a bit... */ 3078 if ((dst_orig->flags & DST_NOXFRM) || 3079 !net->xfrm.policy_count[XFRM_POLICY_OUT]) 3080 goto nopol; 3081 3082 xdst = xfrm_bundle_lookup(net, fl, family, dir, &xflo, if_id); 3083 if (xdst == NULL) 3084 goto nopol; 3085 if (IS_ERR(xdst)) { 3086 err = PTR_ERR(xdst); 3087 goto dropdst; 3088 } 3089 3090 num_pols = xdst->num_pols; 3091 num_xfrms = xdst->num_xfrms; 3092 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols); 3093 route = xdst->route; 3094 } 3095 3096 dst = &xdst->u.dst; 3097 if (route == NULL && num_xfrms > 0) { 3098 /* The only case when xfrm_bundle_lookup() returns a 3099 * bundle with null route, is when the template could 3100 * not be resolved. It means policies are there, but 3101 * bundle could not be created, since we don't yet 3102 * have the xfrm_state's. We need to wait for KM to 3103 * negotiate new SA's or bail out with error.*/ 3104 if (net->xfrm.sysctl_larval_drop) { 3105 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES); 3106 err = -EREMOTE; 3107 goto error; 3108 } 3109 3110 err = -EAGAIN; 3111 3112 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES); 3113 goto error; 3114 } 3115 3116 no_transform: 3117 if (num_pols == 0) 3118 goto nopol; 3119 3120 if ((flags & XFRM_LOOKUP_ICMP) && 3121 !(pols[0]->flags & XFRM_POLICY_ICMP)) { 3122 err = -ENOENT; 3123 goto error; 3124 } 3125 3126 for (i = 0; i < num_pols; i++) 3127 pols[i]->curlft.use_time = ktime_get_real_seconds(); 3128 3129 if (num_xfrms < 0) { 3130 /* Prohibit the flow */ 3131 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK); 3132 err = -EPERM; 3133 goto error; 3134 } else if (num_xfrms > 0) { 3135 /* Flow transformed */ 3136 dst_release(dst_orig); 3137 } else { 3138 /* Flow passes untransformed */ 3139 dst_release(dst); 3140 dst = dst_orig; 3141 } 3142 ok: 3143 xfrm_pols_put(pols, drop_pols); 3144 if (dst && dst->xfrm && 3145 dst->xfrm->props.mode == XFRM_MODE_TUNNEL) 3146 dst->flags |= DST_XFRM_TUNNEL; 3147 return dst; 3148 3149 nopol: 3150 if (!(flags & XFRM_LOOKUP_ICMP)) { 3151 dst = dst_orig; 3152 goto ok; 3153 } 3154 err = -ENOENT; 3155 error: 3156 dst_release(dst); 3157 dropdst: 3158 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF)) 3159 dst_release(dst_orig); 3160 xfrm_pols_put(pols, drop_pols); 3161 return ERR_PTR(err); 3162 } 3163 EXPORT_SYMBOL(xfrm_lookup_with_ifid); 3164 3165 /* Main function: finds/creates a bundle for given flow. 3166 * 3167 * At the moment we eat a raw IP route. Mostly to speed up lookups 3168 * on interfaces with disabled IPsec. 3169 */ 3170 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig, 3171 const struct flowi *fl, const struct sock *sk, 3172 int flags) 3173 { 3174 return xfrm_lookup_with_ifid(net, dst_orig, fl, sk, flags, 0); 3175 } 3176 EXPORT_SYMBOL(xfrm_lookup); 3177 3178 /* Callers of xfrm_lookup_route() must ensure a call to dst_output(). 3179 * Otherwise we may send out blackholed packets. 3180 */ 3181 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig, 3182 const struct flowi *fl, 3183 const struct sock *sk, int flags) 3184 { 3185 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk, 3186 flags | XFRM_LOOKUP_QUEUE | 3187 XFRM_LOOKUP_KEEP_DST_REF); 3188 3189 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE) 3190 return make_blackhole(net, dst_orig->ops->family, dst_orig); 3191 3192 if (IS_ERR(dst)) 3193 dst_release(dst_orig); 3194 3195 return dst; 3196 } 3197 EXPORT_SYMBOL(xfrm_lookup_route); 3198 3199 static inline int 3200 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl) 3201 { 3202 struct sec_path *sp = skb_sec_path(skb); 3203 struct xfrm_state *x; 3204 3205 if (!sp || idx < 0 || idx >= sp->len) 3206 return 0; 3207 x = sp->xvec[idx]; 3208 if (!x->type->reject) 3209 return 0; 3210 return x->type->reject(x, skb, fl); 3211 } 3212 3213 /* When skb is transformed back to its "native" form, we have to 3214 * check policy restrictions. At the moment we make this in maximally 3215 * stupid way. Shame on me. :-) Of course, connected sockets must 3216 * have policy cached at them. 3217 */ 3218 3219 static inline int 3220 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x, 3221 unsigned short family) 3222 { 3223 if (xfrm_state_kern(x)) 3224 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family); 3225 return x->id.proto == tmpl->id.proto && 3226 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) && 3227 (x->props.reqid == tmpl->reqid || !tmpl->reqid) && 3228 x->props.mode == tmpl->mode && 3229 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) || 3230 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) && 3231 !(x->props.mode != XFRM_MODE_TRANSPORT && 3232 xfrm_state_addr_cmp(tmpl, x, family)); 3233 } 3234 3235 /* 3236 * 0 or more than 0 is returned when validation is succeeded (either bypass 3237 * because of optional transport mode, or next index of the mathced secpath 3238 * state with the template. 3239 * -1 is returned when no matching template is found. 3240 * Otherwise "-2 - errored_index" is returned. 3241 */ 3242 static inline int 3243 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start, 3244 unsigned short family) 3245 { 3246 int idx = start; 3247 3248 if (tmpl->optional) { 3249 if (tmpl->mode == XFRM_MODE_TRANSPORT) 3250 return start; 3251 } else 3252 start = -1; 3253 for (; idx < sp->len; idx++) { 3254 if (xfrm_state_ok(tmpl, sp->xvec[idx], family)) 3255 return ++idx; 3256 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) { 3257 if (start == -1) 3258 start = -2-idx; 3259 break; 3260 } 3261 } 3262 return start; 3263 } 3264 3265 static void 3266 decode_session4(struct sk_buff *skb, struct flowi *fl, bool reverse) 3267 { 3268 const struct iphdr *iph = ip_hdr(skb); 3269 int ihl = iph->ihl; 3270 u8 *xprth = skb_network_header(skb) + ihl * 4; 3271 struct flowi4 *fl4 = &fl->u.ip4; 3272 int oif = 0; 3273 3274 if (skb_dst(skb) && skb_dst(skb)->dev) 3275 oif = skb_dst(skb)->dev->ifindex; 3276 3277 memset(fl4, 0, sizeof(struct flowi4)); 3278 fl4->flowi4_mark = skb->mark; 3279 fl4->flowi4_oif = reverse ? skb->skb_iif : oif; 3280 3281 fl4->flowi4_proto = iph->protocol; 3282 fl4->daddr = reverse ? iph->saddr : iph->daddr; 3283 fl4->saddr = reverse ? iph->daddr : iph->saddr; 3284 fl4->flowi4_tos = iph->tos; 3285 3286 if (!ip_is_fragment(iph)) { 3287 switch (iph->protocol) { 3288 case IPPROTO_UDP: 3289 case IPPROTO_UDPLITE: 3290 case IPPROTO_TCP: 3291 case IPPROTO_SCTP: 3292 case IPPROTO_DCCP: 3293 if (xprth + 4 < skb->data || 3294 pskb_may_pull(skb, xprth + 4 - skb->data)) { 3295 __be16 *ports; 3296 3297 xprth = skb_network_header(skb) + ihl * 4; 3298 ports = (__be16 *)xprth; 3299 3300 fl4->fl4_sport = ports[!!reverse]; 3301 fl4->fl4_dport = ports[!reverse]; 3302 } 3303 break; 3304 case IPPROTO_ICMP: 3305 if (xprth + 2 < skb->data || 3306 pskb_may_pull(skb, xprth + 2 - skb->data)) { 3307 u8 *icmp; 3308 3309 xprth = skb_network_header(skb) + ihl * 4; 3310 icmp = xprth; 3311 3312 fl4->fl4_icmp_type = icmp[0]; 3313 fl4->fl4_icmp_code = icmp[1]; 3314 } 3315 break; 3316 case IPPROTO_ESP: 3317 if (xprth + 4 < skb->data || 3318 pskb_may_pull(skb, xprth + 4 - skb->data)) { 3319 __be32 *ehdr; 3320 3321 xprth = skb_network_header(skb) + ihl * 4; 3322 ehdr = (__be32 *)xprth; 3323 3324 fl4->fl4_ipsec_spi = ehdr[0]; 3325 } 3326 break; 3327 case IPPROTO_AH: 3328 if (xprth + 8 < skb->data || 3329 pskb_may_pull(skb, xprth + 8 - skb->data)) { 3330 __be32 *ah_hdr; 3331 3332 xprth = skb_network_header(skb) + ihl * 4; 3333 ah_hdr = (__be32 *)xprth; 3334 3335 fl4->fl4_ipsec_spi = ah_hdr[1]; 3336 } 3337 break; 3338 case IPPROTO_COMP: 3339 if (xprth + 4 < skb->data || 3340 pskb_may_pull(skb, xprth + 4 - skb->data)) { 3341 __be16 *ipcomp_hdr; 3342 3343 xprth = skb_network_header(skb) + ihl * 4; 3344 ipcomp_hdr = (__be16 *)xprth; 3345 3346 fl4->fl4_ipsec_spi = htonl(ntohs(ipcomp_hdr[1])); 3347 } 3348 break; 3349 case IPPROTO_GRE: 3350 if (xprth + 12 < skb->data || 3351 pskb_may_pull(skb, xprth + 12 - skb->data)) { 3352 __be16 *greflags; 3353 __be32 *gre_hdr; 3354 3355 xprth = skb_network_header(skb) + ihl * 4; 3356 greflags = (__be16 *)xprth; 3357 gre_hdr = (__be32 *)xprth; 3358 3359 if (greflags[0] & GRE_KEY) { 3360 if (greflags[0] & GRE_CSUM) 3361 gre_hdr++; 3362 fl4->fl4_gre_key = gre_hdr[1]; 3363 } 3364 } 3365 break; 3366 default: 3367 fl4->fl4_ipsec_spi = 0; 3368 break; 3369 } 3370 } 3371 } 3372 3373 #if IS_ENABLED(CONFIG_IPV6) 3374 static void 3375 decode_session6(struct sk_buff *skb, struct flowi *fl, bool reverse) 3376 { 3377 struct flowi6 *fl6 = &fl->u.ip6; 3378 int onlyproto = 0; 3379 const struct ipv6hdr *hdr = ipv6_hdr(skb); 3380 u32 offset = sizeof(*hdr); 3381 struct ipv6_opt_hdr *exthdr; 3382 const unsigned char *nh = skb_network_header(skb); 3383 u16 nhoff = IP6CB(skb)->nhoff; 3384 int oif = 0; 3385 u8 nexthdr; 3386 3387 if (!nhoff) 3388 nhoff = offsetof(struct ipv6hdr, nexthdr); 3389 3390 nexthdr = nh[nhoff]; 3391 3392 if (skb_dst(skb) && skb_dst(skb)->dev) 3393 oif = skb_dst(skb)->dev->ifindex; 3394 3395 memset(fl6, 0, sizeof(struct flowi6)); 3396 fl6->flowi6_mark = skb->mark; 3397 fl6->flowi6_oif = reverse ? skb->skb_iif : oif; 3398 3399 fl6->daddr = reverse ? hdr->saddr : hdr->daddr; 3400 fl6->saddr = reverse ? hdr->daddr : hdr->saddr; 3401 3402 while (nh + offset + sizeof(*exthdr) < skb->data || 3403 pskb_may_pull(skb, nh + offset + sizeof(*exthdr) - skb->data)) { 3404 nh = skb_network_header(skb); 3405 exthdr = (struct ipv6_opt_hdr *)(nh + offset); 3406 3407 switch (nexthdr) { 3408 case NEXTHDR_FRAGMENT: 3409 onlyproto = 1; 3410 /* fall through */ 3411 case NEXTHDR_ROUTING: 3412 case NEXTHDR_HOP: 3413 case NEXTHDR_DEST: 3414 offset += ipv6_optlen(exthdr); 3415 nexthdr = exthdr->nexthdr; 3416 exthdr = (struct ipv6_opt_hdr *)(nh + offset); 3417 break; 3418 case IPPROTO_UDP: 3419 case IPPROTO_UDPLITE: 3420 case IPPROTO_TCP: 3421 case IPPROTO_SCTP: 3422 case IPPROTO_DCCP: 3423 if (!onlyproto && (nh + offset + 4 < skb->data || 3424 pskb_may_pull(skb, nh + offset + 4 - skb->data))) { 3425 __be16 *ports; 3426 3427 nh = skb_network_header(skb); 3428 ports = (__be16 *)(nh + offset); 3429 fl6->fl6_sport = ports[!!reverse]; 3430 fl6->fl6_dport = ports[!reverse]; 3431 } 3432 fl6->flowi6_proto = nexthdr; 3433 return; 3434 case IPPROTO_ICMPV6: 3435 if (!onlyproto && (nh + offset + 2 < skb->data || 3436 pskb_may_pull(skb, nh + offset + 2 - skb->data))) { 3437 u8 *icmp; 3438 3439 nh = skb_network_header(skb); 3440 icmp = (u8 *)(nh + offset); 3441 fl6->fl6_icmp_type = icmp[0]; 3442 fl6->fl6_icmp_code = icmp[1]; 3443 } 3444 fl6->flowi6_proto = nexthdr; 3445 return; 3446 #if IS_ENABLED(CONFIG_IPV6_MIP6) 3447 case IPPROTO_MH: 3448 offset += ipv6_optlen(exthdr); 3449 if (!onlyproto && (nh + offset + 3 < skb->data || 3450 pskb_may_pull(skb, nh + offset + 3 - skb->data))) { 3451 struct ip6_mh *mh; 3452 3453 nh = skb_network_header(skb); 3454 mh = (struct ip6_mh *)(nh + offset); 3455 fl6->fl6_mh_type = mh->ip6mh_type; 3456 } 3457 fl6->flowi6_proto = nexthdr; 3458 return; 3459 #endif 3460 /* XXX Why are there these headers? */ 3461 case IPPROTO_AH: 3462 case IPPROTO_ESP: 3463 case IPPROTO_COMP: 3464 default: 3465 fl6->fl6_ipsec_spi = 0; 3466 fl6->flowi6_proto = nexthdr; 3467 return; 3468 } 3469 } 3470 } 3471 #endif 3472 3473 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, 3474 unsigned int family, int reverse) 3475 { 3476 switch (family) { 3477 case AF_INET: 3478 decode_session4(skb, fl, reverse); 3479 break; 3480 #if IS_ENABLED(CONFIG_IPV6) 3481 case AF_INET6: 3482 decode_session6(skb, fl, reverse); 3483 break; 3484 #endif 3485 default: 3486 return -EAFNOSUPPORT; 3487 } 3488 3489 return security_xfrm_decode_session(skb, &fl->flowi_secid); 3490 } 3491 EXPORT_SYMBOL(__xfrm_decode_session); 3492 3493 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp) 3494 { 3495 for (; k < sp->len; k++) { 3496 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) { 3497 *idxp = k; 3498 return 1; 3499 } 3500 } 3501 3502 return 0; 3503 } 3504 3505 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 3506 unsigned short family) 3507 { 3508 struct net *net = dev_net(skb->dev); 3509 struct xfrm_policy *pol; 3510 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 3511 int npols = 0; 3512 int xfrm_nr; 3513 int pi; 3514 int reverse; 3515 struct flowi fl; 3516 int xerr_idx = -1; 3517 const struct xfrm_if_cb *ifcb; 3518 struct sec_path *sp; 3519 struct xfrm_if *xi; 3520 u32 if_id = 0; 3521 3522 rcu_read_lock(); 3523 ifcb = xfrm_if_get_cb(); 3524 3525 if (ifcb) { 3526 xi = ifcb->decode_session(skb, family); 3527 if (xi) { 3528 if_id = xi->p.if_id; 3529 net = xi->net; 3530 } 3531 } 3532 rcu_read_unlock(); 3533 3534 reverse = dir & ~XFRM_POLICY_MASK; 3535 dir &= XFRM_POLICY_MASK; 3536 3537 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) { 3538 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR); 3539 return 0; 3540 } 3541 3542 nf_nat_decode_session(skb, &fl, family); 3543 3544 /* First, check used SA against their selectors. */ 3545 sp = skb_sec_path(skb); 3546 if (sp) { 3547 int i; 3548 3549 for (i = sp->len - 1; i >= 0; i--) { 3550 struct xfrm_state *x = sp->xvec[i]; 3551 if (!xfrm_selector_match(&x->sel, &fl, family)) { 3552 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH); 3553 return 0; 3554 } 3555 } 3556 } 3557 3558 pol = NULL; 3559 sk = sk_to_full_sk(sk); 3560 if (sk && sk->sk_policy[dir]) { 3561 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family, if_id); 3562 if (IS_ERR(pol)) { 3563 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3564 return 0; 3565 } 3566 } 3567 3568 if (!pol) 3569 pol = xfrm_policy_lookup(net, &fl, family, dir, if_id); 3570 3571 if (IS_ERR(pol)) { 3572 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3573 return 0; 3574 } 3575 3576 if (!pol) { 3577 if (sp && secpath_has_nontransport(sp, 0, &xerr_idx)) { 3578 xfrm_secpath_reject(xerr_idx, skb, &fl); 3579 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS); 3580 return 0; 3581 } 3582 return 1; 3583 } 3584 3585 pol->curlft.use_time = ktime_get_real_seconds(); 3586 3587 pols[0] = pol; 3588 npols++; 3589 #ifdef CONFIG_XFRM_SUB_POLICY 3590 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) { 3591 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, 3592 &fl, family, 3593 XFRM_POLICY_IN, if_id); 3594 if (pols[1]) { 3595 if (IS_ERR(pols[1])) { 3596 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 3597 return 0; 3598 } 3599 pols[1]->curlft.use_time = ktime_get_real_seconds(); 3600 npols++; 3601 } 3602 } 3603 #endif 3604 3605 if (pol->action == XFRM_POLICY_ALLOW) { 3606 static struct sec_path dummy; 3607 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH]; 3608 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH]; 3609 struct xfrm_tmpl **tpp = tp; 3610 int ti = 0; 3611 int i, k; 3612 3613 sp = skb_sec_path(skb); 3614 if (!sp) 3615 sp = &dummy; 3616 3617 for (pi = 0; pi < npols; pi++) { 3618 if (pols[pi] != pol && 3619 pols[pi]->action != XFRM_POLICY_ALLOW) { 3620 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK); 3621 goto reject; 3622 } 3623 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) { 3624 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR); 3625 goto reject_error; 3626 } 3627 for (i = 0; i < pols[pi]->xfrm_nr; i++) 3628 tpp[ti++] = &pols[pi]->xfrm_vec[i]; 3629 } 3630 xfrm_nr = ti; 3631 if (npols > 1) { 3632 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family); 3633 tpp = stp; 3634 } 3635 3636 /* For each tunnel xfrm, find the first matching tmpl. 3637 * For each tmpl before that, find corresponding xfrm. 3638 * Order is _important_. Later we will implement 3639 * some barriers, but at the moment barriers 3640 * are implied between each two transformations. 3641 */ 3642 for (i = xfrm_nr-1, k = 0; i >= 0; i--) { 3643 k = xfrm_policy_ok(tpp[i], sp, k, family); 3644 if (k < 0) { 3645 if (k < -1) 3646 /* "-2 - errored_index" returned */ 3647 xerr_idx = -(2+k); 3648 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH); 3649 goto reject; 3650 } 3651 } 3652 3653 if (secpath_has_nontransport(sp, k, &xerr_idx)) { 3654 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH); 3655 goto reject; 3656 } 3657 3658 xfrm_pols_put(pols, npols); 3659 return 1; 3660 } 3661 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK); 3662 3663 reject: 3664 xfrm_secpath_reject(xerr_idx, skb, &fl); 3665 reject_error: 3666 xfrm_pols_put(pols, npols); 3667 return 0; 3668 } 3669 EXPORT_SYMBOL(__xfrm_policy_check); 3670 3671 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family) 3672 { 3673 struct net *net = dev_net(skb->dev); 3674 struct flowi fl; 3675 struct dst_entry *dst; 3676 int res = 1; 3677 3678 if (xfrm_decode_session(skb, &fl, family) < 0) { 3679 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR); 3680 return 0; 3681 } 3682 3683 skb_dst_force(skb); 3684 if (!skb_dst(skb)) { 3685 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR); 3686 return 0; 3687 } 3688 3689 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE); 3690 if (IS_ERR(dst)) { 3691 res = 0; 3692 dst = NULL; 3693 } 3694 skb_dst_set(skb, dst); 3695 return res; 3696 } 3697 EXPORT_SYMBOL(__xfrm_route_forward); 3698 3699 /* Optimize later using cookies and generation ids. */ 3700 3701 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie) 3702 { 3703 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete 3704 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to 3705 * get validated by dst_ops->check on every use. We do this 3706 * because when a normal route referenced by an XFRM dst is 3707 * obsoleted we do not go looking around for all parent 3708 * referencing XFRM dsts so that we can invalidate them. It 3709 * is just too much work. Instead we make the checks here on 3710 * every use. For example: 3711 * 3712 * XFRM dst A --> IPv4 dst X 3713 * 3714 * X is the "xdst->route" of A (X is also the "dst->path" of A 3715 * in this example). If X is marked obsolete, "A" will not 3716 * notice. That's what we are validating here via the 3717 * stale_bundle() check. 3718 * 3719 * When a dst is removed from the fib tree, DST_OBSOLETE_DEAD will 3720 * be marked on it. 3721 * This will force stale_bundle() to fail on any xdst bundle with 3722 * this dst linked in it. 3723 */ 3724 if (dst->obsolete < 0 && !stale_bundle(dst)) 3725 return dst; 3726 3727 return NULL; 3728 } 3729 3730 static int stale_bundle(struct dst_entry *dst) 3731 { 3732 return !xfrm_bundle_ok((struct xfrm_dst *)dst); 3733 } 3734 3735 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev) 3736 { 3737 while ((dst = xfrm_dst_child(dst)) && dst->xfrm && dst->dev == dev) { 3738 dst->dev = dev_net(dev)->loopback_dev; 3739 dev_hold(dst->dev); 3740 dev_put(dev); 3741 } 3742 } 3743 EXPORT_SYMBOL(xfrm_dst_ifdown); 3744 3745 static void xfrm_link_failure(struct sk_buff *skb) 3746 { 3747 /* Impossible. Such dst must be popped before reaches point of failure. */ 3748 } 3749 3750 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst) 3751 { 3752 if (dst) { 3753 if (dst->obsolete) { 3754 dst_release(dst); 3755 dst = NULL; 3756 } 3757 } 3758 return dst; 3759 } 3760 3761 static void xfrm_init_pmtu(struct xfrm_dst **bundle, int nr) 3762 { 3763 while (nr--) { 3764 struct xfrm_dst *xdst = bundle[nr]; 3765 u32 pmtu, route_mtu_cached; 3766 struct dst_entry *dst; 3767 3768 dst = &xdst->u.dst; 3769 pmtu = dst_mtu(xfrm_dst_child(dst)); 3770 xdst->child_mtu_cached = pmtu; 3771 3772 pmtu = xfrm_state_mtu(dst->xfrm, pmtu); 3773 3774 route_mtu_cached = dst_mtu(xdst->route); 3775 xdst->route_mtu_cached = route_mtu_cached; 3776 3777 if (pmtu > route_mtu_cached) 3778 pmtu = route_mtu_cached; 3779 3780 dst_metric_set(dst, RTAX_MTU, pmtu); 3781 } 3782 } 3783 3784 /* Check that the bundle accepts the flow and its components are 3785 * still valid. 3786 */ 3787 3788 static int xfrm_bundle_ok(struct xfrm_dst *first) 3789 { 3790 struct xfrm_dst *bundle[XFRM_MAX_DEPTH]; 3791 struct dst_entry *dst = &first->u.dst; 3792 struct xfrm_dst *xdst; 3793 int start_from, nr; 3794 u32 mtu; 3795 3796 if (!dst_check(xfrm_dst_path(dst), ((struct xfrm_dst *)dst)->path_cookie) || 3797 (dst->dev && !netif_running(dst->dev))) 3798 return 0; 3799 3800 if (dst->flags & DST_XFRM_QUEUE) 3801 return 1; 3802 3803 start_from = nr = 0; 3804 do { 3805 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 3806 3807 if (dst->xfrm->km.state != XFRM_STATE_VALID) 3808 return 0; 3809 if (xdst->xfrm_genid != dst->xfrm->genid) 3810 return 0; 3811 if (xdst->num_pols > 0 && 3812 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid)) 3813 return 0; 3814 3815 bundle[nr++] = xdst; 3816 3817 mtu = dst_mtu(xfrm_dst_child(dst)); 3818 if (xdst->child_mtu_cached != mtu) { 3819 start_from = nr; 3820 xdst->child_mtu_cached = mtu; 3821 } 3822 3823 if (!dst_check(xdst->route, xdst->route_cookie)) 3824 return 0; 3825 mtu = dst_mtu(xdst->route); 3826 if (xdst->route_mtu_cached != mtu) { 3827 start_from = nr; 3828 xdst->route_mtu_cached = mtu; 3829 } 3830 3831 dst = xfrm_dst_child(dst); 3832 } while (dst->xfrm); 3833 3834 if (likely(!start_from)) 3835 return 1; 3836 3837 xdst = bundle[start_from - 1]; 3838 mtu = xdst->child_mtu_cached; 3839 while (start_from--) { 3840 dst = &xdst->u.dst; 3841 3842 mtu = xfrm_state_mtu(dst->xfrm, mtu); 3843 if (mtu > xdst->route_mtu_cached) 3844 mtu = xdst->route_mtu_cached; 3845 dst_metric_set(dst, RTAX_MTU, mtu); 3846 if (!start_from) 3847 break; 3848 3849 xdst = bundle[start_from - 1]; 3850 xdst->child_mtu_cached = mtu; 3851 } 3852 3853 return 1; 3854 } 3855 3856 static unsigned int xfrm_default_advmss(const struct dst_entry *dst) 3857 { 3858 return dst_metric_advmss(xfrm_dst_path(dst)); 3859 } 3860 3861 static unsigned int xfrm_mtu(const struct dst_entry *dst) 3862 { 3863 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 3864 3865 return mtu ? : dst_mtu(xfrm_dst_path(dst)); 3866 } 3867 3868 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst, 3869 const void *daddr) 3870 { 3871 while (dst->xfrm) { 3872 const struct xfrm_state *xfrm = dst->xfrm; 3873 3874 dst = xfrm_dst_child(dst); 3875 3876 if (xfrm->props.mode == XFRM_MODE_TRANSPORT) 3877 continue; 3878 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR) 3879 daddr = xfrm->coaddr; 3880 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR)) 3881 daddr = &xfrm->id.daddr; 3882 } 3883 return daddr; 3884 } 3885 3886 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst, 3887 struct sk_buff *skb, 3888 const void *daddr) 3889 { 3890 const struct dst_entry *path = xfrm_dst_path(dst); 3891 3892 if (!skb) 3893 daddr = xfrm_get_dst_nexthop(dst, daddr); 3894 return path->ops->neigh_lookup(path, skb, daddr); 3895 } 3896 3897 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr) 3898 { 3899 const struct dst_entry *path = xfrm_dst_path(dst); 3900 3901 daddr = xfrm_get_dst_nexthop(dst, daddr); 3902 path->ops->confirm_neigh(path, daddr); 3903 } 3904 3905 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family) 3906 { 3907 int err = 0; 3908 3909 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo))) 3910 return -EAFNOSUPPORT; 3911 3912 spin_lock(&xfrm_policy_afinfo_lock); 3913 if (unlikely(xfrm_policy_afinfo[family] != NULL)) 3914 err = -EEXIST; 3915 else { 3916 struct dst_ops *dst_ops = afinfo->dst_ops; 3917 if (likely(dst_ops->kmem_cachep == NULL)) 3918 dst_ops->kmem_cachep = xfrm_dst_cache; 3919 if (likely(dst_ops->check == NULL)) 3920 dst_ops->check = xfrm_dst_check; 3921 if (likely(dst_ops->default_advmss == NULL)) 3922 dst_ops->default_advmss = xfrm_default_advmss; 3923 if (likely(dst_ops->mtu == NULL)) 3924 dst_ops->mtu = xfrm_mtu; 3925 if (likely(dst_ops->negative_advice == NULL)) 3926 dst_ops->negative_advice = xfrm_negative_advice; 3927 if (likely(dst_ops->link_failure == NULL)) 3928 dst_ops->link_failure = xfrm_link_failure; 3929 if (likely(dst_ops->neigh_lookup == NULL)) 3930 dst_ops->neigh_lookup = xfrm_neigh_lookup; 3931 if (likely(!dst_ops->confirm_neigh)) 3932 dst_ops->confirm_neigh = xfrm_confirm_neigh; 3933 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo); 3934 } 3935 spin_unlock(&xfrm_policy_afinfo_lock); 3936 3937 return err; 3938 } 3939 EXPORT_SYMBOL(xfrm_policy_register_afinfo); 3940 3941 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo) 3942 { 3943 struct dst_ops *dst_ops = afinfo->dst_ops; 3944 int i; 3945 3946 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) { 3947 if (xfrm_policy_afinfo[i] != afinfo) 3948 continue; 3949 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL); 3950 break; 3951 } 3952 3953 synchronize_rcu(); 3954 3955 dst_ops->kmem_cachep = NULL; 3956 dst_ops->check = NULL; 3957 dst_ops->negative_advice = NULL; 3958 dst_ops->link_failure = NULL; 3959 } 3960 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo); 3961 3962 void xfrm_if_register_cb(const struct xfrm_if_cb *ifcb) 3963 { 3964 spin_lock(&xfrm_if_cb_lock); 3965 rcu_assign_pointer(xfrm_if_cb, ifcb); 3966 spin_unlock(&xfrm_if_cb_lock); 3967 } 3968 EXPORT_SYMBOL(xfrm_if_register_cb); 3969 3970 void xfrm_if_unregister_cb(void) 3971 { 3972 RCU_INIT_POINTER(xfrm_if_cb, NULL); 3973 synchronize_rcu(); 3974 } 3975 EXPORT_SYMBOL(xfrm_if_unregister_cb); 3976 3977 #ifdef CONFIG_XFRM_STATISTICS 3978 static int __net_init xfrm_statistics_init(struct net *net) 3979 { 3980 int rv; 3981 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib); 3982 if (!net->mib.xfrm_statistics) 3983 return -ENOMEM; 3984 rv = xfrm_proc_init(net); 3985 if (rv < 0) 3986 free_percpu(net->mib.xfrm_statistics); 3987 return rv; 3988 } 3989 3990 static void xfrm_statistics_fini(struct net *net) 3991 { 3992 xfrm_proc_fini(net); 3993 free_percpu(net->mib.xfrm_statistics); 3994 } 3995 #else 3996 static int __net_init xfrm_statistics_init(struct net *net) 3997 { 3998 return 0; 3999 } 4000 4001 static void xfrm_statistics_fini(struct net *net) 4002 { 4003 } 4004 #endif 4005 4006 static int __net_init xfrm_policy_init(struct net *net) 4007 { 4008 unsigned int hmask, sz; 4009 int dir, err; 4010 4011 if (net_eq(net, &init_net)) { 4012 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache", 4013 sizeof(struct xfrm_dst), 4014 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, 4015 NULL); 4016 err = rhashtable_init(&xfrm_policy_inexact_table, 4017 &xfrm_pol_inexact_params); 4018 BUG_ON(err); 4019 } 4020 4021 hmask = 8 - 1; 4022 sz = (hmask+1) * sizeof(struct hlist_head); 4023 4024 net->xfrm.policy_byidx = xfrm_hash_alloc(sz); 4025 if (!net->xfrm.policy_byidx) 4026 goto out_byidx; 4027 net->xfrm.policy_idx_hmask = hmask; 4028 4029 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 4030 struct xfrm_policy_hash *htab; 4031 4032 net->xfrm.policy_count[dir] = 0; 4033 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0; 4034 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]); 4035 4036 htab = &net->xfrm.policy_bydst[dir]; 4037 htab->table = xfrm_hash_alloc(sz); 4038 if (!htab->table) 4039 goto out_bydst; 4040 htab->hmask = hmask; 4041 htab->dbits4 = 32; 4042 htab->sbits4 = 32; 4043 htab->dbits6 = 128; 4044 htab->sbits6 = 128; 4045 } 4046 net->xfrm.policy_hthresh.lbits4 = 32; 4047 net->xfrm.policy_hthresh.rbits4 = 32; 4048 net->xfrm.policy_hthresh.lbits6 = 128; 4049 net->xfrm.policy_hthresh.rbits6 = 128; 4050 4051 seqlock_init(&net->xfrm.policy_hthresh.lock); 4052 4053 INIT_LIST_HEAD(&net->xfrm.policy_all); 4054 INIT_LIST_HEAD(&net->xfrm.inexact_bins); 4055 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize); 4056 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild); 4057 return 0; 4058 4059 out_bydst: 4060 for (dir--; dir >= 0; dir--) { 4061 struct xfrm_policy_hash *htab; 4062 4063 htab = &net->xfrm.policy_bydst[dir]; 4064 xfrm_hash_free(htab->table, sz); 4065 } 4066 xfrm_hash_free(net->xfrm.policy_byidx, sz); 4067 out_byidx: 4068 return -ENOMEM; 4069 } 4070 4071 static void xfrm_policy_fini(struct net *net) 4072 { 4073 struct xfrm_pol_inexact_bin *b, *t; 4074 unsigned int sz; 4075 int dir; 4076 4077 flush_work(&net->xfrm.policy_hash_work); 4078 #ifdef CONFIG_XFRM_SUB_POLICY 4079 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false); 4080 #endif 4081 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false); 4082 4083 WARN_ON(!list_empty(&net->xfrm.policy_all)); 4084 4085 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 4086 struct xfrm_policy_hash *htab; 4087 4088 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir])); 4089 4090 htab = &net->xfrm.policy_bydst[dir]; 4091 sz = (htab->hmask + 1) * sizeof(struct hlist_head); 4092 WARN_ON(!hlist_empty(htab->table)); 4093 xfrm_hash_free(htab->table, sz); 4094 } 4095 4096 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head); 4097 WARN_ON(!hlist_empty(net->xfrm.policy_byidx)); 4098 xfrm_hash_free(net->xfrm.policy_byidx, sz); 4099 4100 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 4101 list_for_each_entry_safe(b, t, &net->xfrm.inexact_bins, inexact_bins) 4102 __xfrm_policy_inexact_prune_bin(b, true); 4103 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 4104 } 4105 4106 static int __net_init xfrm_net_init(struct net *net) 4107 { 4108 int rv; 4109 4110 /* Initialize the per-net locks here */ 4111 spin_lock_init(&net->xfrm.xfrm_state_lock); 4112 spin_lock_init(&net->xfrm.xfrm_policy_lock); 4113 mutex_init(&net->xfrm.xfrm_cfg_mutex); 4114 4115 rv = xfrm_statistics_init(net); 4116 if (rv < 0) 4117 goto out_statistics; 4118 rv = xfrm_state_init(net); 4119 if (rv < 0) 4120 goto out_state; 4121 rv = xfrm_policy_init(net); 4122 if (rv < 0) 4123 goto out_policy; 4124 rv = xfrm_sysctl_init(net); 4125 if (rv < 0) 4126 goto out_sysctl; 4127 4128 return 0; 4129 4130 out_sysctl: 4131 xfrm_policy_fini(net); 4132 out_policy: 4133 xfrm_state_fini(net); 4134 out_state: 4135 xfrm_statistics_fini(net); 4136 out_statistics: 4137 return rv; 4138 } 4139 4140 static void __net_exit xfrm_net_exit(struct net *net) 4141 { 4142 xfrm_sysctl_fini(net); 4143 xfrm_policy_fini(net); 4144 xfrm_state_fini(net); 4145 xfrm_statistics_fini(net); 4146 } 4147 4148 static struct pernet_operations __net_initdata xfrm_net_ops = { 4149 .init = xfrm_net_init, 4150 .exit = xfrm_net_exit, 4151 }; 4152 4153 void __init xfrm_init(void) 4154 { 4155 register_pernet_subsys(&xfrm_net_ops); 4156 xfrm_dev_init(); 4157 seqcount_init(&xfrm_policy_hash_generation); 4158 xfrm_input_init(); 4159 4160 RCU_INIT_POINTER(xfrm_if_cb, NULL); 4161 synchronize_rcu(); 4162 } 4163 4164 #ifdef CONFIG_AUDITSYSCALL 4165 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp, 4166 struct audit_buffer *audit_buf) 4167 { 4168 struct xfrm_sec_ctx *ctx = xp->security; 4169 struct xfrm_selector *sel = &xp->selector; 4170 4171 if (ctx) 4172 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s", 4173 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str); 4174 4175 switch (sel->family) { 4176 case AF_INET: 4177 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4); 4178 if (sel->prefixlen_s != 32) 4179 audit_log_format(audit_buf, " src_prefixlen=%d", 4180 sel->prefixlen_s); 4181 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4); 4182 if (sel->prefixlen_d != 32) 4183 audit_log_format(audit_buf, " dst_prefixlen=%d", 4184 sel->prefixlen_d); 4185 break; 4186 case AF_INET6: 4187 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6); 4188 if (sel->prefixlen_s != 128) 4189 audit_log_format(audit_buf, " src_prefixlen=%d", 4190 sel->prefixlen_s); 4191 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6); 4192 if (sel->prefixlen_d != 128) 4193 audit_log_format(audit_buf, " dst_prefixlen=%d", 4194 sel->prefixlen_d); 4195 break; 4196 } 4197 } 4198 4199 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid) 4200 { 4201 struct audit_buffer *audit_buf; 4202 4203 audit_buf = xfrm_audit_start("SPD-add"); 4204 if (audit_buf == NULL) 4205 return; 4206 xfrm_audit_helper_usrinfo(task_valid, audit_buf); 4207 audit_log_format(audit_buf, " res=%u", result); 4208 xfrm_audit_common_policyinfo(xp, audit_buf); 4209 audit_log_end(audit_buf); 4210 } 4211 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add); 4212 4213 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result, 4214 bool task_valid) 4215 { 4216 struct audit_buffer *audit_buf; 4217 4218 audit_buf = xfrm_audit_start("SPD-delete"); 4219 if (audit_buf == NULL) 4220 return; 4221 xfrm_audit_helper_usrinfo(task_valid, audit_buf); 4222 audit_log_format(audit_buf, " res=%u", result); 4223 xfrm_audit_common_policyinfo(xp, audit_buf); 4224 audit_log_end(audit_buf); 4225 } 4226 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete); 4227 #endif 4228 4229 #ifdef CONFIG_XFRM_MIGRATE 4230 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp, 4231 const struct xfrm_selector *sel_tgt) 4232 { 4233 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) { 4234 if (sel_tgt->family == sel_cmp->family && 4235 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr, 4236 sel_cmp->family) && 4237 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr, 4238 sel_cmp->family) && 4239 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d && 4240 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) { 4241 return true; 4242 } 4243 } else { 4244 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) { 4245 return true; 4246 } 4247 } 4248 return false; 4249 } 4250 4251 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel, 4252 u8 dir, u8 type, struct net *net) 4253 { 4254 struct xfrm_policy *pol, *ret = NULL; 4255 struct hlist_head *chain; 4256 u32 priority = ~0U; 4257 4258 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 4259 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir); 4260 hlist_for_each_entry(pol, chain, bydst) { 4261 if (xfrm_migrate_selector_match(sel, &pol->selector) && 4262 pol->type == type) { 4263 ret = pol; 4264 priority = ret->priority; 4265 break; 4266 } 4267 } 4268 chain = &net->xfrm.policy_inexact[dir]; 4269 hlist_for_each_entry(pol, chain, bydst_inexact_list) { 4270 if ((pol->priority >= priority) && ret) 4271 break; 4272 4273 if (xfrm_migrate_selector_match(sel, &pol->selector) && 4274 pol->type == type) { 4275 ret = pol; 4276 break; 4277 } 4278 } 4279 4280 xfrm_pol_hold(ret); 4281 4282 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 4283 4284 return ret; 4285 } 4286 4287 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t) 4288 { 4289 int match = 0; 4290 4291 if (t->mode == m->mode && t->id.proto == m->proto && 4292 (m->reqid == 0 || t->reqid == m->reqid)) { 4293 switch (t->mode) { 4294 case XFRM_MODE_TUNNEL: 4295 case XFRM_MODE_BEET: 4296 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr, 4297 m->old_family) && 4298 xfrm_addr_equal(&t->saddr, &m->old_saddr, 4299 m->old_family)) { 4300 match = 1; 4301 } 4302 break; 4303 case XFRM_MODE_TRANSPORT: 4304 /* in case of transport mode, template does not store 4305 any IP addresses, hence we just compare mode and 4306 protocol */ 4307 match = 1; 4308 break; 4309 default: 4310 break; 4311 } 4312 } 4313 return match; 4314 } 4315 4316 /* update endpoint address(es) of template(s) */ 4317 static int xfrm_policy_migrate(struct xfrm_policy *pol, 4318 struct xfrm_migrate *m, int num_migrate) 4319 { 4320 struct xfrm_migrate *mp; 4321 int i, j, n = 0; 4322 4323 write_lock_bh(&pol->lock); 4324 if (unlikely(pol->walk.dead)) { 4325 /* target policy has been deleted */ 4326 write_unlock_bh(&pol->lock); 4327 return -ENOENT; 4328 } 4329 4330 for (i = 0; i < pol->xfrm_nr; i++) { 4331 for (j = 0, mp = m; j < num_migrate; j++, mp++) { 4332 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i])) 4333 continue; 4334 n++; 4335 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL && 4336 pol->xfrm_vec[i].mode != XFRM_MODE_BEET) 4337 continue; 4338 /* update endpoints */ 4339 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr, 4340 sizeof(pol->xfrm_vec[i].id.daddr)); 4341 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr, 4342 sizeof(pol->xfrm_vec[i].saddr)); 4343 pol->xfrm_vec[i].encap_family = mp->new_family; 4344 /* flush bundles */ 4345 atomic_inc(&pol->genid); 4346 } 4347 } 4348 4349 write_unlock_bh(&pol->lock); 4350 4351 if (!n) 4352 return -ENODATA; 4353 4354 return 0; 4355 } 4356 4357 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate) 4358 { 4359 int i, j; 4360 4361 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH) 4362 return -EINVAL; 4363 4364 for (i = 0; i < num_migrate; i++) { 4365 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) || 4366 xfrm_addr_any(&m[i].new_saddr, m[i].new_family)) 4367 return -EINVAL; 4368 4369 /* check if there is any duplicated entry */ 4370 for (j = i + 1; j < num_migrate; j++) { 4371 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr, 4372 sizeof(m[i].old_daddr)) && 4373 !memcmp(&m[i].old_saddr, &m[j].old_saddr, 4374 sizeof(m[i].old_saddr)) && 4375 m[i].proto == m[j].proto && 4376 m[i].mode == m[j].mode && 4377 m[i].reqid == m[j].reqid && 4378 m[i].old_family == m[j].old_family) 4379 return -EINVAL; 4380 } 4381 } 4382 4383 return 0; 4384 } 4385 4386 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type, 4387 struct xfrm_migrate *m, int num_migrate, 4388 struct xfrm_kmaddress *k, struct net *net, 4389 struct xfrm_encap_tmpl *encap) 4390 { 4391 int i, err, nx_cur = 0, nx_new = 0; 4392 struct xfrm_policy *pol = NULL; 4393 struct xfrm_state *x, *xc; 4394 struct xfrm_state *x_cur[XFRM_MAX_DEPTH]; 4395 struct xfrm_state *x_new[XFRM_MAX_DEPTH]; 4396 struct xfrm_migrate *mp; 4397 4398 /* Stage 0 - sanity checks */ 4399 if ((err = xfrm_migrate_check(m, num_migrate)) < 0) 4400 goto out; 4401 4402 if (dir >= XFRM_POLICY_MAX) { 4403 err = -EINVAL; 4404 goto out; 4405 } 4406 4407 /* Stage 1 - find policy */ 4408 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) { 4409 err = -ENOENT; 4410 goto out; 4411 } 4412 4413 /* Stage 2 - find and update state(s) */ 4414 for (i = 0, mp = m; i < num_migrate; i++, mp++) { 4415 if ((x = xfrm_migrate_state_find(mp, net))) { 4416 x_cur[nx_cur] = x; 4417 nx_cur++; 4418 xc = xfrm_state_migrate(x, mp, encap); 4419 if (xc) { 4420 x_new[nx_new] = xc; 4421 nx_new++; 4422 } else { 4423 err = -ENODATA; 4424 goto restore_state; 4425 } 4426 } 4427 } 4428 4429 /* Stage 3 - update policy */ 4430 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0) 4431 goto restore_state; 4432 4433 /* Stage 4 - delete old state(s) */ 4434 if (nx_cur) { 4435 xfrm_states_put(x_cur, nx_cur); 4436 xfrm_states_delete(x_cur, nx_cur); 4437 } 4438 4439 /* Stage 5 - announce */ 4440 km_migrate(sel, dir, type, m, num_migrate, k, encap); 4441 4442 xfrm_pol_put(pol); 4443 4444 return 0; 4445 out: 4446 return err; 4447 4448 restore_state: 4449 if (pol) 4450 xfrm_pol_put(pol); 4451 if (nx_cur) 4452 xfrm_states_put(x_cur, nx_cur); 4453 if (nx_new) 4454 xfrm_states_delete(x_new, nx_new); 4455 4456 return err; 4457 } 4458 EXPORT_SYMBOL(xfrm_migrate); 4459 #endif 4460