1 /* 2 * xfrm_policy.c 3 * 4 * Changes: 5 * Mitsuru KANDA @USAGI 6 * Kazunori MIYAZAWA @USAGI 7 * Kunihiro Ishiguro <kunihiro@ipinfusion.com> 8 * IPv6 support 9 * Kazunori MIYAZAWA @USAGI 10 * YOSHIFUJI Hideaki 11 * Split up af-specific portion 12 * Derek Atkins <derek@ihtfp.com> Add the post_input processor 13 * 14 */ 15 16 #include <linux/err.h> 17 #include <linux/slab.h> 18 #include <linux/kmod.h> 19 #include <linux/list.h> 20 #include <linux/spinlock.h> 21 #include <linux/workqueue.h> 22 #include <linux/notifier.h> 23 #include <linux/netdevice.h> 24 #include <linux/netfilter.h> 25 #include <linux/module.h> 26 #include <linux/cache.h> 27 #include <linux/audit.h> 28 #include <net/dst.h> 29 #include <net/flow.h> 30 #include <net/xfrm.h> 31 #include <net/ip.h> 32 #ifdef CONFIG_XFRM_STATISTICS 33 #include <net/snmp.h> 34 #endif 35 36 #include "xfrm_hash.h" 37 38 #define XFRM_QUEUE_TMO_MIN ((unsigned)(HZ/10)) 39 #define XFRM_QUEUE_TMO_MAX ((unsigned)(60*HZ)) 40 #define XFRM_MAX_QUEUE_LEN 100 41 42 struct xfrm_flo { 43 struct dst_entry *dst_orig; 44 u8 flags; 45 }; 46 47 static DEFINE_SPINLOCK(xfrm_policy_afinfo_lock); 48 static struct xfrm_policy_afinfo const __rcu *xfrm_policy_afinfo[AF_INET6 + 1] 49 __read_mostly; 50 51 static struct kmem_cache *xfrm_dst_cache __read_mostly; 52 static __read_mostly seqcount_t xfrm_policy_hash_generation; 53 54 static void xfrm_init_pmtu(struct dst_entry *dst); 55 static int stale_bundle(struct dst_entry *dst); 56 static int xfrm_bundle_ok(struct xfrm_dst *xdst); 57 static void xfrm_policy_queue_process(unsigned long arg); 58 59 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir); 60 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol, 61 int dir); 62 63 static inline bool xfrm_pol_hold_rcu(struct xfrm_policy *policy) 64 { 65 return atomic_inc_not_zero(&policy->refcnt); 66 } 67 68 static inline bool 69 __xfrm4_selector_match(const struct xfrm_selector *sel, const struct flowi *fl) 70 { 71 const struct flowi4 *fl4 = &fl->u.ip4; 72 73 return addr4_match(fl4->daddr, sel->daddr.a4, sel->prefixlen_d) && 74 addr4_match(fl4->saddr, sel->saddr.a4, sel->prefixlen_s) && 75 !((xfrm_flowi_dport(fl, &fl4->uli) ^ sel->dport) & sel->dport_mask) && 76 !((xfrm_flowi_sport(fl, &fl4->uli) ^ sel->sport) & sel->sport_mask) && 77 (fl4->flowi4_proto == sel->proto || !sel->proto) && 78 (fl4->flowi4_oif == sel->ifindex || !sel->ifindex); 79 } 80 81 static inline bool 82 __xfrm6_selector_match(const struct xfrm_selector *sel, const struct flowi *fl) 83 { 84 const struct flowi6 *fl6 = &fl->u.ip6; 85 86 return addr_match(&fl6->daddr, &sel->daddr, sel->prefixlen_d) && 87 addr_match(&fl6->saddr, &sel->saddr, sel->prefixlen_s) && 88 !((xfrm_flowi_dport(fl, &fl6->uli) ^ sel->dport) & sel->dport_mask) && 89 !((xfrm_flowi_sport(fl, &fl6->uli) ^ sel->sport) & sel->sport_mask) && 90 (fl6->flowi6_proto == sel->proto || !sel->proto) && 91 (fl6->flowi6_oif == sel->ifindex || !sel->ifindex); 92 } 93 94 bool xfrm_selector_match(const struct xfrm_selector *sel, const struct flowi *fl, 95 unsigned short family) 96 { 97 switch (family) { 98 case AF_INET: 99 return __xfrm4_selector_match(sel, fl); 100 case AF_INET6: 101 return __xfrm6_selector_match(sel, fl); 102 } 103 return false; 104 } 105 106 static const struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family) 107 { 108 const struct xfrm_policy_afinfo *afinfo; 109 110 if (unlikely(family >= ARRAY_SIZE(xfrm_policy_afinfo))) 111 return NULL; 112 rcu_read_lock(); 113 afinfo = rcu_dereference(xfrm_policy_afinfo[family]); 114 if (unlikely(!afinfo)) 115 rcu_read_unlock(); 116 return afinfo; 117 } 118 119 static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, 120 int tos, int oif, 121 const xfrm_address_t *saddr, 122 const xfrm_address_t *daddr, 123 int family) 124 { 125 const struct xfrm_policy_afinfo *afinfo; 126 struct dst_entry *dst; 127 128 afinfo = xfrm_policy_get_afinfo(family); 129 if (unlikely(afinfo == NULL)) 130 return ERR_PTR(-EAFNOSUPPORT); 131 132 dst = afinfo->dst_lookup(net, tos, oif, saddr, daddr); 133 134 rcu_read_unlock(); 135 136 return dst; 137 } 138 139 static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, 140 int tos, int oif, 141 xfrm_address_t *prev_saddr, 142 xfrm_address_t *prev_daddr, 143 int family) 144 { 145 struct net *net = xs_net(x); 146 xfrm_address_t *saddr = &x->props.saddr; 147 xfrm_address_t *daddr = &x->id.daddr; 148 struct dst_entry *dst; 149 150 if (x->type->flags & XFRM_TYPE_LOCAL_COADDR) { 151 saddr = x->coaddr; 152 daddr = prev_daddr; 153 } 154 if (x->type->flags & XFRM_TYPE_REMOTE_COADDR) { 155 saddr = prev_saddr; 156 daddr = x->coaddr; 157 } 158 159 dst = __xfrm_dst_lookup(net, tos, oif, saddr, daddr, family); 160 161 if (!IS_ERR(dst)) { 162 if (prev_saddr != saddr) 163 memcpy(prev_saddr, saddr, sizeof(*prev_saddr)); 164 if (prev_daddr != daddr) 165 memcpy(prev_daddr, daddr, sizeof(*prev_daddr)); 166 } 167 168 return dst; 169 } 170 171 static inline unsigned long make_jiffies(long secs) 172 { 173 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ) 174 return MAX_SCHEDULE_TIMEOUT-1; 175 else 176 return secs*HZ; 177 } 178 179 static void xfrm_policy_timer(unsigned long data) 180 { 181 struct xfrm_policy *xp = (struct xfrm_policy *)data; 182 unsigned long now = get_seconds(); 183 long next = LONG_MAX; 184 int warn = 0; 185 int dir; 186 187 read_lock(&xp->lock); 188 189 if (unlikely(xp->walk.dead)) 190 goto out; 191 192 dir = xfrm_policy_id2dir(xp->index); 193 194 if (xp->lft.hard_add_expires_seconds) { 195 long tmo = xp->lft.hard_add_expires_seconds + 196 xp->curlft.add_time - now; 197 if (tmo <= 0) 198 goto expired; 199 if (tmo < next) 200 next = tmo; 201 } 202 if (xp->lft.hard_use_expires_seconds) { 203 long tmo = xp->lft.hard_use_expires_seconds + 204 (xp->curlft.use_time ? : xp->curlft.add_time) - now; 205 if (tmo <= 0) 206 goto expired; 207 if (tmo < next) 208 next = tmo; 209 } 210 if (xp->lft.soft_add_expires_seconds) { 211 long tmo = xp->lft.soft_add_expires_seconds + 212 xp->curlft.add_time - now; 213 if (tmo <= 0) { 214 warn = 1; 215 tmo = XFRM_KM_TIMEOUT; 216 } 217 if (tmo < next) 218 next = tmo; 219 } 220 if (xp->lft.soft_use_expires_seconds) { 221 long tmo = xp->lft.soft_use_expires_seconds + 222 (xp->curlft.use_time ? : xp->curlft.add_time) - now; 223 if (tmo <= 0) { 224 warn = 1; 225 tmo = XFRM_KM_TIMEOUT; 226 } 227 if (tmo < next) 228 next = tmo; 229 } 230 231 if (warn) 232 km_policy_expired(xp, dir, 0, 0); 233 if (next != LONG_MAX && 234 !mod_timer(&xp->timer, jiffies + make_jiffies(next))) 235 xfrm_pol_hold(xp); 236 237 out: 238 read_unlock(&xp->lock); 239 xfrm_pol_put(xp); 240 return; 241 242 expired: 243 read_unlock(&xp->lock); 244 if (!xfrm_policy_delete(xp, dir)) 245 km_policy_expired(xp, dir, 1, 0); 246 xfrm_pol_put(xp); 247 } 248 249 static struct flow_cache_object *xfrm_policy_flo_get(struct flow_cache_object *flo) 250 { 251 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo); 252 253 if (unlikely(pol->walk.dead)) 254 flo = NULL; 255 else 256 xfrm_pol_hold(pol); 257 258 return flo; 259 } 260 261 static int xfrm_policy_flo_check(struct flow_cache_object *flo) 262 { 263 struct xfrm_policy *pol = container_of(flo, struct xfrm_policy, flo); 264 265 return !pol->walk.dead; 266 } 267 268 static void xfrm_policy_flo_delete(struct flow_cache_object *flo) 269 { 270 xfrm_pol_put(container_of(flo, struct xfrm_policy, flo)); 271 } 272 273 static const struct flow_cache_ops xfrm_policy_fc_ops = { 274 .get = xfrm_policy_flo_get, 275 .check = xfrm_policy_flo_check, 276 .delete = xfrm_policy_flo_delete, 277 }; 278 279 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2 280 * SPD calls. 281 */ 282 283 struct xfrm_policy *xfrm_policy_alloc(struct net *net, gfp_t gfp) 284 { 285 struct xfrm_policy *policy; 286 287 policy = kzalloc(sizeof(struct xfrm_policy), gfp); 288 289 if (policy) { 290 write_pnet(&policy->xp_net, net); 291 INIT_LIST_HEAD(&policy->walk.all); 292 INIT_HLIST_NODE(&policy->bydst); 293 INIT_HLIST_NODE(&policy->byidx); 294 rwlock_init(&policy->lock); 295 atomic_set(&policy->refcnt, 1); 296 skb_queue_head_init(&policy->polq.hold_queue); 297 setup_timer(&policy->timer, xfrm_policy_timer, 298 (unsigned long)policy); 299 setup_timer(&policy->polq.hold_timer, xfrm_policy_queue_process, 300 (unsigned long)policy); 301 policy->flo.ops = &xfrm_policy_fc_ops; 302 } 303 return policy; 304 } 305 EXPORT_SYMBOL(xfrm_policy_alloc); 306 307 static void xfrm_policy_destroy_rcu(struct rcu_head *head) 308 { 309 struct xfrm_policy *policy = container_of(head, struct xfrm_policy, rcu); 310 311 security_xfrm_policy_free(policy->security); 312 kfree(policy); 313 } 314 315 /* Destroy xfrm_policy: descendant resources must be released to this moment. */ 316 317 void xfrm_policy_destroy(struct xfrm_policy *policy) 318 { 319 BUG_ON(!policy->walk.dead); 320 321 if (del_timer(&policy->timer) || del_timer(&policy->polq.hold_timer)) 322 BUG(); 323 324 call_rcu(&policy->rcu, xfrm_policy_destroy_rcu); 325 } 326 EXPORT_SYMBOL(xfrm_policy_destroy); 327 328 /* Rule must be locked. Release descendant resources, announce 329 * entry dead. The rule must be unlinked from lists to the moment. 330 */ 331 332 static void xfrm_policy_kill(struct xfrm_policy *policy) 333 { 334 policy->walk.dead = 1; 335 336 atomic_inc(&policy->genid); 337 338 if (del_timer(&policy->polq.hold_timer)) 339 xfrm_pol_put(policy); 340 skb_queue_purge(&policy->polq.hold_queue); 341 342 if (del_timer(&policy->timer)) 343 xfrm_pol_put(policy); 344 345 xfrm_pol_put(policy); 346 } 347 348 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024; 349 350 static inline unsigned int idx_hash(struct net *net, u32 index) 351 { 352 return __idx_hash(index, net->xfrm.policy_idx_hmask); 353 } 354 355 /* calculate policy hash thresholds */ 356 static void __get_hash_thresh(struct net *net, 357 unsigned short family, int dir, 358 u8 *dbits, u8 *sbits) 359 { 360 switch (family) { 361 case AF_INET: 362 *dbits = net->xfrm.policy_bydst[dir].dbits4; 363 *sbits = net->xfrm.policy_bydst[dir].sbits4; 364 break; 365 366 case AF_INET6: 367 *dbits = net->xfrm.policy_bydst[dir].dbits6; 368 *sbits = net->xfrm.policy_bydst[dir].sbits6; 369 break; 370 371 default: 372 *dbits = 0; 373 *sbits = 0; 374 } 375 } 376 377 static struct hlist_head *policy_hash_bysel(struct net *net, 378 const struct xfrm_selector *sel, 379 unsigned short family, int dir) 380 { 381 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 382 unsigned int hash; 383 u8 dbits; 384 u8 sbits; 385 386 __get_hash_thresh(net, family, dir, &dbits, &sbits); 387 hash = __sel_hash(sel, family, hmask, dbits, sbits); 388 389 if (hash == hmask + 1) 390 return &net->xfrm.policy_inexact[dir]; 391 392 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table, 393 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash; 394 } 395 396 static struct hlist_head *policy_hash_direct(struct net *net, 397 const xfrm_address_t *daddr, 398 const xfrm_address_t *saddr, 399 unsigned short family, int dir) 400 { 401 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 402 unsigned int hash; 403 u8 dbits; 404 u8 sbits; 405 406 __get_hash_thresh(net, family, dir, &dbits, &sbits); 407 hash = __addr_hash(daddr, saddr, family, hmask, dbits, sbits); 408 409 return rcu_dereference_check(net->xfrm.policy_bydst[dir].table, 410 lockdep_is_held(&net->xfrm.xfrm_policy_lock)) + hash; 411 } 412 413 static void xfrm_dst_hash_transfer(struct net *net, 414 struct hlist_head *list, 415 struct hlist_head *ndsttable, 416 unsigned int nhashmask, 417 int dir) 418 { 419 struct hlist_node *tmp, *entry0 = NULL; 420 struct xfrm_policy *pol; 421 unsigned int h0 = 0; 422 u8 dbits; 423 u8 sbits; 424 425 redo: 426 hlist_for_each_entry_safe(pol, tmp, list, bydst) { 427 unsigned int h; 428 429 __get_hash_thresh(net, pol->family, dir, &dbits, &sbits); 430 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr, 431 pol->family, nhashmask, dbits, sbits); 432 if (!entry0) { 433 hlist_del_rcu(&pol->bydst); 434 hlist_add_head_rcu(&pol->bydst, ndsttable + h); 435 h0 = h; 436 } else { 437 if (h != h0) 438 continue; 439 hlist_del_rcu(&pol->bydst); 440 hlist_add_behind_rcu(&pol->bydst, entry0); 441 } 442 entry0 = &pol->bydst; 443 } 444 if (!hlist_empty(list)) { 445 entry0 = NULL; 446 goto redo; 447 } 448 } 449 450 static void xfrm_idx_hash_transfer(struct hlist_head *list, 451 struct hlist_head *nidxtable, 452 unsigned int nhashmask) 453 { 454 struct hlist_node *tmp; 455 struct xfrm_policy *pol; 456 457 hlist_for_each_entry_safe(pol, tmp, list, byidx) { 458 unsigned int h; 459 460 h = __idx_hash(pol->index, nhashmask); 461 hlist_add_head(&pol->byidx, nidxtable+h); 462 } 463 } 464 465 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask) 466 { 467 return ((old_hmask + 1) << 1) - 1; 468 } 469 470 static void xfrm_bydst_resize(struct net *net, int dir) 471 { 472 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 473 unsigned int nhashmask = xfrm_new_hash_mask(hmask); 474 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head); 475 struct hlist_head *ndst = xfrm_hash_alloc(nsize); 476 struct hlist_head *odst; 477 int i; 478 479 if (!ndst) 480 return; 481 482 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 483 write_seqcount_begin(&xfrm_policy_hash_generation); 484 485 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table, 486 lockdep_is_held(&net->xfrm.xfrm_policy_lock)); 487 488 odst = rcu_dereference_protected(net->xfrm.policy_bydst[dir].table, 489 lockdep_is_held(&net->xfrm.xfrm_policy_lock)); 490 491 for (i = hmask; i >= 0; i--) 492 xfrm_dst_hash_transfer(net, odst + i, ndst, nhashmask, dir); 493 494 rcu_assign_pointer(net->xfrm.policy_bydst[dir].table, ndst); 495 net->xfrm.policy_bydst[dir].hmask = nhashmask; 496 497 write_seqcount_end(&xfrm_policy_hash_generation); 498 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 499 500 synchronize_rcu(); 501 502 xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head)); 503 } 504 505 static void xfrm_byidx_resize(struct net *net, int total) 506 { 507 unsigned int hmask = net->xfrm.policy_idx_hmask; 508 unsigned int nhashmask = xfrm_new_hash_mask(hmask); 509 unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head); 510 struct hlist_head *oidx = net->xfrm.policy_byidx; 511 struct hlist_head *nidx = xfrm_hash_alloc(nsize); 512 int i; 513 514 if (!nidx) 515 return; 516 517 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 518 519 for (i = hmask; i >= 0; i--) 520 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask); 521 522 net->xfrm.policy_byidx = nidx; 523 net->xfrm.policy_idx_hmask = nhashmask; 524 525 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 526 527 xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head)); 528 } 529 530 static inline int xfrm_bydst_should_resize(struct net *net, int dir, int *total) 531 { 532 unsigned int cnt = net->xfrm.policy_count[dir]; 533 unsigned int hmask = net->xfrm.policy_bydst[dir].hmask; 534 535 if (total) 536 *total += cnt; 537 538 if ((hmask + 1) < xfrm_policy_hashmax && 539 cnt > hmask) 540 return 1; 541 542 return 0; 543 } 544 545 static inline int xfrm_byidx_should_resize(struct net *net, int total) 546 { 547 unsigned int hmask = net->xfrm.policy_idx_hmask; 548 549 if ((hmask + 1) < xfrm_policy_hashmax && 550 total > hmask) 551 return 1; 552 553 return 0; 554 } 555 556 void xfrm_spd_getinfo(struct net *net, struct xfrmk_spdinfo *si) 557 { 558 si->incnt = net->xfrm.policy_count[XFRM_POLICY_IN]; 559 si->outcnt = net->xfrm.policy_count[XFRM_POLICY_OUT]; 560 si->fwdcnt = net->xfrm.policy_count[XFRM_POLICY_FWD]; 561 si->inscnt = net->xfrm.policy_count[XFRM_POLICY_IN+XFRM_POLICY_MAX]; 562 si->outscnt = net->xfrm.policy_count[XFRM_POLICY_OUT+XFRM_POLICY_MAX]; 563 si->fwdscnt = net->xfrm.policy_count[XFRM_POLICY_FWD+XFRM_POLICY_MAX]; 564 si->spdhcnt = net->xfrm.policy_idx_hmask; 565 si->spdhmcnt = xfrm_policy_hashmax; 566 } 567 EXPORT_SYMBOL(xfrm_spd_getinfo); 568 569 static DEFINE_MUTEX(hash_resize_mutex); 570 static void xfrm_hash_resize(struct work_struct *work) 571 { 572 struct net *net = container_of(work, struct net, xfrm.policy_hash_work); 573 int dir, total; 574 575 mutex_lock(&hash_resize_mutex); 576 577 total = 0; 578 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 579 if (xfrm_bydst_should_resize(net, dir, &total)) 580 xfrm_bydst_resize(net, dir); 581 } 582 if (xfrm_byidx_should_resize(net, total)) 583 xfrm_byidx_resize(net, total); 584 585 mutex_unlock(&hash_resize_mutex); 586 } 587 588 static void xfrm_hash_rebuild(struct work_struct *work) 589 { 590 struct net *net = container_of(work, struct net, 591 xfrm.policy_hthresh.work); 592 unsigned int hmask; 593 struct xfrm_policy *pol; 594 struct xfrm_policy *policy; 595 struct hlist_head *chain; 596 struct hlist_head *odst; 597 struct hlist_node *newpos; 598 int i; 599 int dir; 600 unsigned seq; 601 u8 lbits4, rbits4, lbits6, rbits6; 602 603 mutex_lock(&hash_resize_mutex); 604 605 /* read selector prefixlen thresholds */ 606 do { 607 seq = read_seqbegin(&net->xfrm.policy_hthresh.lock); 608 609 lbits4 = net->xfrm.policy_hthresh.lbits4; 610 rbits4 = net->xfrm.policy_hthresh.rbits4; 611 lbits6 = net->xfrm.policy_hthresh.lbits6; 612 rbits6 = net->xfrm.policy_hthresh.rbits6; 613 } while (read_seqretry(&net->xfrm.policy_hthresh.lock, seq)); 614 615 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 616 617 /* reset the bydst and inexact table in all directions */ 618 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 619 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]); 620 hmask = net->xfrm.policy_bydst[dir].hmask; 621 odst = net->xfrm.policy_bydst[dir].table; 622 for (i = hmask; i >= 0; i--) 623 INIT_HLIST_HEAD(odst + i); 624 if ((dir & XFRM_POLICY_MASK) == XFRM_POLICY_OUT) { 625 /* dir out => dst = remote, src = local */ 626 net->xfrm.policy_bydst[dir].dbits4 = rbits4; 627 net->xfrm.policy_bydst[dir].sbits4 = lbits4; 628 net->xfrm.policy_bydst[dir].dbits6 = rbits6; 629 net->xfrm.policy_bydst[dir].sbits6 = lbits6; 630 } else { 631 /* dir in/fwd => dst = local, src = remote */ 632 net->xfrm.policy_bydst[dir].dbits4 = lbits4; 633 net->xfrm.policy_bydst[dir].sbits4 = rbits4; 634 net->xfrm.policy_bydst[dir].dbits6 = lbits6; 635 net->xfrm.policy_bydst[dir].sbits6 = rbits6; 636 } 637 } 638 639 /* re-insert all policies by order of creation */ 640 list_for_each_entry_reverse(policy, &net->xfrm.policy_all, walk.all) { 641 if (xfrm_policy_id2dir(policy->index) >= XFRM_POLICY_MAX) { 642 /* skip socket policies */ 643 continue; 644 } 645 newpos = NULL; 646 chain = policy_hash_bysel(net, &policy->selector, 647 policy->family, 648 xfrm_policy_id2dir(policy->index)); 649 hlist_for_each_entry(pol, chain, bydst) { 650 if (policy->priority >= pol->priority) 651 newpos = &pol->bydst; 652 else 653 break; 654 } 655 if (newpos) 656 hlist_add_behind(&policy->bydst, newpos); 657 else 658 hlist_add_head(&policy->bydst, chain); 659 } 660 661 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 662 663 mutex_unlock(&hash_resize_mutex); 664 } 665 666 void xfrm_policy_hash_rebuild(struct net *net) 667 { 668 schedule_work(&net->xfrm.policy_hthresh.work); 669 } 670 EXPORT_SYMBOL(xfrm_policy_hash_rebuild); 671 672 /* Generate new index... KAME seems to generate them ordered by cost 673 * of an absolute inpredictability of ordering of rules. This will not pass. */ 674 static u32 xfrm_gen_index(struct net *net, int dir, u32 index) 675 { 676 static u32 idx_generator; 677 678 for (;;) { 679 struct hlist_head *list; 680 struct xfrm_policy *p; 681 u32 idx; 682 int found; 683 684 if (!index) { 685 idx = (idx_generator | dir); 686 idx_generator += 8; 687 } else { 688 idx = index; 689 index = 0; 690 } 691 692 if (idx == 0) 693 idx = 8; 694 list = net->xfrm.policy_byidx + idx_hash(net, idx); 695 found = 0; 696 hlist_for_each_entry(p, list, byidx) { 697 if (p->index == idx) { 698 found = 1; 699 break; 700 } 701 } 702 if (!found) 703 return idx; 704 } 705 } 706 707 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2) 708 { 709 u32 *p1 = (u32 *) s1; 710 u32 *p2 = (u32 *) s2; 711 int len = sizeof(struct xfrm_selector) / sizeof(u32); 712 int i; 713 714 for (i = 0; i < len; i++) { 715 if (p1[i] != p2[i]) 716 return 1; 717 } 718 719 return 0; 720 } 721 722 static void xfrm_policy_requeue(struct xfrm_policy *old, 723 struct xfrm_policy *new) 724 { 725 struct xfrm_policy_queue *pq = &old->polq; 726 struct sk_buff_head list; 727 728 if (skb_queue_empty(&pq->hold_queue)) 729 return; 730 731 __skb_queue_head_init(&list); 732 733 spin_lock_bh(&pq->hold_queue.lock); 734 skb_queue_splice_init(&pq->hold_queue, &list); 735 if (del_timer(&pq->hold_timer)) 736 xfrm_pol_put(old); 737 spin_unlock_bh(&pq->hold_queue.lock); 738 739 pq = &new->polq; 740 741 spin_lock_bh(&pq->hold_queue.lock); 742 skb_queue_splice(&list, &pq->hold_queue); 743 pq->timeout = XFRM_QUEUE_TMO_MIN; 744 if (!mod_timer(&pq->hold_timer, jiffies)) 745 xfrm_pol_hold(new); 746 spin_unlock_bh(&pq->hold_queue.lock); 747 } 748 749 static bool xfrm_policy_mark_match(struct xfrm_policy *policy, 750 struct xfrm_policy *pol) 751 { 752 u32 mark = policy->mark.v & policy->mark.m; 753 754 if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m) 755 return true; 756 757 if ((mark & pol->mark.m) == pol->mark.v && 758 policy->priority == pol->priority) 759 return true; 760 761 return false; 762 } 763 764 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl) 765 { 766 struct net *net = xp_net(policy); 767 struct xfrm_policy *pol; 768 struct xfrm_policy *delpol; 769 struct hlist_head *chain; 770 struct hlist_node *newpos; 771 772 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 773 chain = policy_hash_bysel(net, &policy->selector, policy->family, dir); 774 delpol = NULL; 775 newpos = NULL; 776 hlist_for_each_entry(pol, chain, bydst) { 777 if (pol->type == policy->type && 778 !selector_cmp(&pol->selector, &policy->selector) && 779 xfrm_policy_mark_match(policy, pol) && 780 xfrm_sec_ctx_match(pol->security, policy->security) && 781 !WARN_ON(delpol)) { 782 if (excl) { 783 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 784 return -EEXIST; 785 } 786 delpol = pol; 787 if (policy->priority > pol->priority) 788 continue; 789 } else if (policy->priority >= pol->priority) { 790 newpos = &pol->bydst; 791 continue; 792 } 793 if (delpol) 794 break; 795 } 796 if (newpos) 797 hlist_add_behind(&policy->bydst, newpos); 798 else 799 hlist_add_head(&policy->bydst, chain); 800 __xfrm_policy_link(policy, dir); 801 atomic_inc(&net->xfrm.flow_cache_genid); 802 803 /* After previous checking, family can either be AF_INET or AF_INET6 */ 804 if (policy->family == AF_INET) 805 rt_genid_bump_ipv4(net); 806 else 807 rt_genid_bump_ipv6(net); 808 809 if (delpol) { 810 xfrm_policy_requeue(delpol, policy); 811 __xfrm_policy_unlink(delpol, dir); 812 } 813 policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir, policy->index); 814 hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index)); 815 policy->curlft.add_time = get_seconds(); 816 policy->curlft.use_time = 0; 817 if (!mod_timer(&policy->timer, jiffies + HZ)) 818 xfrm_pol_hold(policy); 819 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 820 821 if (delpol) 822 xfrm_policy_kill(delpol); 823 else if (xfrm_bydst_should_resize(net, dir, NULL)) 824 schedule_work(&net->xfrm.policy_hash_work); 825 826 return 0; 827 } 828 EXPORT_SYMBOL(xfrm_policy_insert); 829 830 struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u32 mark, u8 type, 831 int dir, struct xfrm_selector *sel, 832 struct xfrm_sec_ctx *ctx, int delete, 833 int *err) 834 { 835 struct xfrm_policy *pol, *ret; 836 struct hlist_head *chain; 837 838 *err = 0; 839 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 840 chain = policy_hash_bysel(net, sel, sel->family, dir); 841 ret = NULL; 842 hlist_for_each_entry(pol, chain, bydst) { 843 if (pol->type == type && 844 (mark & pol->mark.m) == pol->mark.v && 845 !selector_cmp(sel, &pol->selector) && 846 xfrm_sec_ctx_match(ctx, pol->security)) { 847 xfrm_pol_hold(pol); 848 if (delete) { 849 *err = security_xfrm_policy_delete( 850 pol->security); 851 if (*err) { 852 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 853 return pol; 854 } 855 __xfrm_policy_unlink(pol, dir); 856 } 857 ret = pol; 858 break; 859 } 860 } 861 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 862 863 if (ret && delete) 864 xfrm_policy_kill(ret); 865 return ret; 866 } 867 EXPORT_SYMBOL(xfrm_policy_bysel_ctx); 868 869 struct xfrm_policy *xfrm_policy_byid(struct net *net, u32 mark, u8 type, 870 int dir, u32 id, int delete, int *err) 871 { 872 struct xfrm_policy *pol, *ret; 873 struct hlist_head *chain; 874 875 *err = -ENOENT; 876 if (xfrm_policy_id2dir(id) != dir) 877 return NULL; 878 879 *err = 0; 880 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 881 chain = net->xfrm.policy_byidx + idx_hash(net, id); 882 ret = NULL; 883 hlist_for_each_entry(pol, chain, byidx) { 884 if (pol->type == type && pol->index == id && 885 (mark & pol->mark.m) == pol->mark.v) { 886 xfrm_pol_hold(pol); 887 if (delete) { 888 *err = security_xfrm_policy_delete( 889 pol->security); 890 if (*err) { 891 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 892 return pol; 893 } 894 __xfrm_policy_unlink(pol, dir); 895 } 896 ret = pol; 897 break; 898 } 899 } 900 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 901 902 if (ret && delete) 903 xfrm_policy_kill(ret); 904 return ret; 905 } 906 EXPORT_SYMBOL(xfrm_policy_byid); 907 908 #ifdef CONFIG_SECURITY_NETWORK_XFRM 909 static inline int 910 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid) 911 { 912 int dir, err = 0; 913 914 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 915 struct xfrm_policy *pol; 916 int i; 917 918 hlist_for_each_entry(pol, 919 &net->xfrm.policy_inexact[dir], bydst) { 920 if (pol->type != type) 921 continue; 922 err = security_xfrm_policy_delete(pol->security); 923 if (err) { 924 xfrm_audit_policy_delete(pol, 0, task_valid); 925 return err; 926 } 927 } 928 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) { 929 hlist_for_each_entry(pol, 930 net->xfrm.policy_bydst[dir].table + i, 931 bydst) { 932 if (pol->type != type) 933 continue; 934 err = security_xfrm_policy_delete( 935 pol->security); 936 if (err) { 937 xfrm_audit_policy_delete(pol, 0, 938 task_valid); 939 return err; 940 } 941 } 942 } 943 } 944 return err; 945 } 946 #else 947 static inline int 948 xfrm_policy_flush_secctx_check(struct net *net, u8 type, bool task_valid) 949 { 950 return 0; 951 } 952 #endif 953 954 int xfrm_policy_flush(struct net *net, u8 type, bool task_valid) 955 { 956 int dir, err = 0, cnt = 0; 957 958 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 959 960 err = xfrm_policy_flush_secctx_check(net, type, task_valid); 961 if (err) 962 goto out; 963 964 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 965 struct xfrm_policy *pol; 966 int i; 967 968 again1: 969 hlist_for_each_entry(pol, 970 &net->xfrm.policy_inexact[dir], bydst) { 971 if (pol->type != type) 972 continue; 973 __xfrm_policy_unlink(pol, dir); 974 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 975 cnt++; 976 977 xfrm_audit_policy_delete(pol, 1, task_valid); 978 979 xfrm_policy_kill(pol); 980 981 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 982 goto again1; 983 } 984 985 for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) { 986 again2: 987 hlist_for_each_entry(pol, 988 net->xfrm.policy_bydst[dir].table + i, 989 bydst) { 990 if (pol->type != type) 991 continue; 992 __xfrm_policy_unlink(pol, dir); 993 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 994 cnt++; 995 996 xfrm_audit_policy_delete(pol, 1, task_valid); 997 xfrm_policy_kill(pol); 998 999 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1000 goto again2; 1001 } 1002 } 1003 1004 } 1005 if (!cnt) 1006 err = -ESRCH; 1007 out: 1008 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1009 1010 if (cnt) 1011 xfrm_garbage_collect(net); 1012 1013 return err; 1014 } 1015 EXPORT_SYMBOL(xfrm_policy_flush); 1016 1017 int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk, 1018 int (*func)(struct xfrm_policy *, int, int, void*), 1019 void *data) 1020 { 1021 struct xfrm_policy *pol; 1022 struct xfrm_policy_walk_entry *x; 1023 int error = 0; 1024 1025 if (walk->type >= XFRM_POLICY_TYPE_MAX && 1026 walk->type != XFRM_POLICY_TYPE_ANY) 1027 return -EINVAL; 1028 1029 if (list_empty(&walk->walk.all) && walk->seq != 0) 1030 return 0; 1031 1032 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1033 if (list_empty(&walk->walk.all)) 1034 x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all); 1035 else 1036 x = list_first_entry(&walk->walk.all, 1037 struct xfrm_policy_walk_entry, all); 1038 1039 list_for_each_entry_from(x, &net->xfrm.policy_all, all) { 1040 if (x->dead) 1041 continue; 1042 pol = container_of(x, struct xfrm_policy, walk); 1043 if (walk->type != XFRM_POLICY_TYPE_ANY && 1044 walk->type != pol->type) 1045 continue; 1046 error = func(pol, xfrm_policy_id2dir(pol->index), 1047 walk->seq, data); 1048 if (error) { 1049 list_move_tail(&walk->walk.all, &x->all); 1050 goto out; 1051 } 1052 walk->seq++; 1053 } 1054 if (walk->seq == 0) { 1055 error = -ENOENT; 1056 goto out; 1057 } 1058 list_del_init(&walk->walk.all); 1059 out: 1060 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1061 return error; 1062 } 1063 EXPORT_SYMBOL(xfrm_policy_walk); 1064 1065 void xfrm_policy_walk_init(struct xfrm_policy_walk *walk, u8 type) 1066 { 1067 INIT_LIST_HEAD(&walk->walk.all); 1068 walk->walk.dead = 1; 1069 walk->type = type; 1070 walk->seq = 0; 1071 } 1072 EXPORT_SYMBOL(xfrm_policy_walk_init); 1073 1074 void xfrm_policy_walk_done(struct xfrm_policy_walk *walk, struct net *net) 1075 { 1076 if (list_empty(&walk->walk.all)) 1077 return; 1078 1079 spin_lock_bh(&net->xfrm.xfrm_policy_lock); /*FIXME where is net? */ 1080 list_del(&walk->walk.all); 1081 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1082 } 1083 EXPORT_SYMBOL(xfrm_policy_walk_done); 1084 1085 /* 1086 * Find policy to apply to this flow. 1087 * 1088 * Returns 0 if policy found, else an -errno. 1089 */ 1090 static int xfrm_policy_match(const struct xfrm_policy *pol, 1091 const struct flowi *fl, 1092 u8 type, u16 family, int dir) 1093 { 1094 const struct xfrm_selector *sel = &pol->selector; 1095 int ret = -ESRCH; 1096 bool match; 1097 1098 if (pol->family != family || 1099 (fl->flowi_mark & pol->mark.m) != pol->mark.v || 1100 pol->type != type) 1101 return ret; 1102 1103 match = xfrm_selector_match(sel, fl, family); 1104 if (match) 1105 ret = security_xfrm_policy_lookup(pol->security, fl->flowi_secid, 1106 dir); 1107 1108 return ret; 1109 } 1110 1111 static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type, 1112 const struct flowi *fl, 1113 u16 family, u8 dir) 1114 { 1115 int err; 1116 struct xfrm_policy *pol, *ret; 1117 const xfrm_address_t *daddr, *saddr; 1118 struct hlist_head *chain; 1119 unsigned int sequence; 1120 u32 priority; 1121 1122 daddr = xfrm_flowi_daddr(fl, family); 1123 saddr = xfrm_flowi_saddr(fl, family); 1124 if (unlikely(!daddr || !saddr)) 1125 return NULL; 1126 1127 rcu_read_lock(); 1128 retry: 1129 do { 1130 sequence = read_seqcount_begin(&xfrm_policy_hash_generation); 1131 chain = policy_hash_direct(net, daddr, saddr, family, dir); 1132 } while (read_seqcount_retry(&xfrm_policy_hash_generation, sequence)); 1133 1134 priority = ~0U; 1135 ret = NULL; 1136 hlist_for_each_entry_rcu(pol, chain, bydst) { 1137 err = xfrm_policy_match(pol, fl, type, family, dir); 1138 if (err) { 1139 if (err == -ESRCH) 1140 continue; 1141 else { 1142 ret = ERR_PTR(err); 1143 goto fail; 1144 } 1145 } else { 1146 ret = pol; 1147 priority = ret->priority; 1148 break; 1149 } 1150 } 1151 chain = &net->xfrm.policy_inexact[dir]; 1152 hlist_for_each_entry_rcu(pol, chain, bydst) { 1153 if ((pol->priority >= priority) && ret) 1154 break; 1155 1156 err = xfrm_policy_match(pol, fl, type, family, dir); 1157 if (err) { 1158 if (err == -ESRCH) 1159 continue; 1160 else { 1161 ret = ERR_PTR(err); 1162 goto fail; 1163 } 1164 } else { 1165 ret = pol; 1166 break; 1167 } 1168 } 1169 1170 if (read_seqcount_retry(&xfrm_policy_hash_generation, sequence)) 1171 goto retry; 1172 1173 if (ret && !xfrm_pol_hold_rcu(ret)) 1174 goto retry; 1175 fail: 1176 rcu_read_unlock(); 1177 1178 return ret; 1179 } 1180 1181 static struct xfrm_policy * 1182 __xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir) 1183 { 1184 #ifdef CONFIG_XFRM_SUB_POLICY 1185 struct xfrm_policy *pol; 1186 1187 pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir); 1188 if (pol != NULL) 1189 return pol; 1190 #endif 1191 return xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir); 1192 } 1193 1194 static int flow_to_policy_dir(int dir) 1195 { 1196 if (XFRM_POLICY_IN == FLOW_DIR_IN && 1197 XFRM_POLICY_OUT == FLOW_DIR_OUT && 1198 XFRM_POLICY_FWD == FLOW_DIR_FWD) 1199 return dir; 1200 1201 switch (dir) { 1202 default: 1203 case FLOW_DIR_IN: 1204 return XFRM_POLICY_IN; 1205 case FLOW_DIR_OUT: 1206 return XFRM_POLICY_OUT; 1207 case FLOW_DIR_FWD: 1208 return XFRM_POLICY_FWD; 1209 } 1210 } 1211 1212 static struct flow_cache_object * 1213 xfrm_policy_lookup(struct net *net, const struct flowi *fl, u16 family, 1214 u8 dir, struct flow_cache_object *old_obj, void *ctx) 1215 { 1216 struct xfrm_policy *pol; 1217 1218 if (old_obj) 1219 xfrm_pol_put(container_of(old_obj, struct xfrm_policy, flo)); 1220 1221 pol = __xfrm_policy_lookup(net, fl, family, flow_to_policy_dir(dir)); 1222 if (IS_ERR_OR_NULL(pol)) 1223 return ERR_CAST(pol); 1224 1225 /* Resolver returns two references: 1226 * one for cache and one for caller of flow_cache_lookup() */ 1227 xfrm_pol_hold(pol); 1228 1229 return &pol->flo; 1230 } 1231 1232 static inline int policy_to_flow_dir(int dir) 1233 { 1234 if (XFRM_POLICY_IN == FLOW_DIR_IN && 1235 XFRM_POLICY_OUT == FLOW_DIR_OUT && 1236 XFRM_POLICY_FWD == FLOW_DIR_FWD) 1237 return dir; 1238 switch (dir) { 1239 default: 1240 case XFRM_POLICY_IN: 1241 return FLOW_DIR_IN; 1242 case XFRM_POLICY_OUT: 1243 return FLOW_DIR_OUT; 1244 case XFRM_POLICY_FWD: 1245 return FLOW_DIR_FWD; 1246 } 1247 } 1248 1249 static struct xfrm_policy *xfrm_sk_policy_lookup(const struct sock *sk, int dir, 1250 const struct flowi *fl, u16 family) 1251 { 1252 struct xfrm_policy *pol; 1253 1254 rcu_read_lock(); 1255 again: 1256 pol = rcu_dereference(sk->sk_policy[dir]); 1257 if (pol != NULL) { 1258 bool match = xfrm_selector_match(&pol->selector, fl, family); 1259 int err = 0; 1260 1261 if (match) { 1262 if ((sk->sk_mark & pol->mark.m) != pol->mark.v) { 1263 pol = NULL; 1264 goto out; 1265 } 1266 err = security_xfrm_policy_lookup(pol->security, 1267 fl->flowi_secid, 1268 policy_to_flow_dir(dir)); 1269 if (!err) { 1270 if (!xfrm_pol_hold_rcu(pol)) 1271 goto again; 1272 } else if (err == -ESRCH) { 1273 pol = NULL; 1274 } else { 1275 pol = ERR_PTR(err); 1276 } 1277 } else 1278 pol = NULL; 1279 } 1280 out: 1281 rcu_read_unlock(); 1282 return pol; 1283 } 1284 1285 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir) 1286 { 1287 struct net *net = xp_net(pol); 1288 1289 list_add(&pol->walk.all, &net->xfrm.policy_all); 1290 net->xfrm.policy_count[dir]++; 1291 xfrm_pol_hold(pol); 1292 } 1293 1294 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol, 1295 int dir) 1296 { 1297 struct net *net = xp_net(pol); 1298 1299 if (list_empty(&pol->walk.all)) 1300 return NULL; 1301 1302 /* Socket policies are not hashed. */ 1303 if (!hlist_unhashed(&pol->bydst)) { 1304 hlist_del_rcu(&pol->bydst); 1305 hlist_del(&pol->byidx); 1306 } 1307 1308 list_del_init(&pol->walk.all); 1309 net->xfrm.policy_count[dir]--; 1310 1311 return pol; 1312 } 1313 1314 static void xfrm_sk_policy_link(struct xfrm_policy *pol, int dir) 1315 { 1316 __xfrm_policy_link(pol, XFRM_POLICY_MAX + dir); 1317 } 1318 1319 static void xfrm_sk_policy_unlink(struct xfrm_policy *pol, int dir) 1320 { 1321 __xfrm_policy_unlink(pol, XFRM_POLICY_MAX + dir); 1322 } 1323 1324 int xfrm_policy_delete(struct xfrm_policy *pol, int dir) 1325 { 1326 struct net *net = xp_net(pol); 1327 1328 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1329 pol = __xfrm_policy_unlink(pol, dir); 1330 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1331 if (pol) { 1332 xfrm_policy_kill(pol); 1333 return 0; 1334 } 1335 return -ENOENT; 1336 } 1337 EXPORT_SYMBOL(xfrm_policy_delete); 1338 1339 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol) 1340 { 1341 struct net *net = xp_net(pol); 1342 struct xfrm_policy *old_pol; 1343 1344 #ifdef CONFIG_XFRM_SUB_POLICY 1345 if (pol && pol->type != XFRM_POLICY_TYPE_MAIN) 1346 return -EINVAL; 1347 #endif 1348 1349 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1350 old_pol = rcu_dereference_protected(sk->sk_policy[dir], 1351 lockdep_is_held(&net->xfrm.xfrm_policy_lock)); 1352 if (pol) { 1353 pol->curlft.add_time = get_seconds(); 1354 pol->index = xfrm_gen_index(net, XFRM_POLICY_MAX+dir, 0); 1355 xfrm_sk_policy_link(pol, dir); 1356 } 1357 rcu_assign_pointer(sk->sk_policy[dir], pol); 1358 if (old_pol) { 1359 if (pol) 1360 xfrm_policy_requeue(old_pol, pol); 1361 1362 /* Unlinking succeeds always. This is the only function 1363 * allowed to delete or replace socket policy. 1364 */ 1365 xfrm_sk_policy_unlink(old_pol, dir); 1366 } 1367 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1368 1369 if (old_pol) { 1370 xfrm_policy_kill(old_pol); 1371 } 1372 return 0; 1373 } 1374 1375 static struct xfrm_policy *clone_policy(const struct xfrm_policy *old, int dir) 1376 { 1377 struct xfrm_policy *newp = xfrm_policy_alloc(xp_net(old), GFP_ATOMIC); 1378 struct net *net = xp_net(old); 1379 1380 if (newp) { 1381 newp->selector = old->selector; 1382 if (security_xfrm_policy_clone(old->security, 1383 &newp->security)) { 1384 kfree(newp); 1385 return NULL; /* ENOMEM */ 1386 } 1387 newp->lft = old->lft; 1388 newp->curlft = old->curlft; 1389 newp->mark = old->mark; 1390 newp->action = old->action; 1391 newp->flags = old->flags; 1392 newp->xfrm_nr = old->xfrm_nr; 1393 newp->index = old->index; 1394 newp->type = old->type; 1395 memcpy(newp->xfrm_vec, old->xfrm_vec, 1396 newp->xfrm_nr*sizeof(struct xfrm_tmpl)); 1397 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 1398 xfrm_sk_policy_link(newp, dir); 1399 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 1400 xfrm_pol_put(newp); 1401 } 1402 return newp; 1403 } 1404 1405 int __xfrm_sk_clone_policy(struct sock *sk, const struct sock *osk) 1406 { 1407 const struct xfrm_policy *p; 1408 struct xfrm_policy *np; 1409 int i, ret = 0; 1410 1411 rcu_read_lock(); 1412 for (i = 0; i < 2; i++) { 1413 p = rcu_dereference(osk->sk_policy[i]); 1414 if (p) { 1415 np = clone_policy(p, i); 1416 if (unlikely(!np)) { 1417 ret = -ENOMEM; 1418 break; 1419 } 1420 rcu_assign_pointer(sk->sk_policy[i], np); 1421 } 1422 } 1423 rcu_read_unlock(); 1424 return ret; 1425 } 1426 1427 static int 1428 xfrm_get_saddr(struct net *net, int oif, xfrm_address_t *local, 1429 xfrm_address_t *remote, unsigned short family) 1430 { 1431 int err; 1432 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 1433 1434 if (unlikely(afinfo == NULL)) 1435 return -EINVAL; 1436 err = afinfo->get_saddr(net, oif, local, remote); 1437 rcu_read_unlock(); 1438 return err; 1439 } 1440 1441 /* Resolve list of templates for the flow, given policy. */ 1442 1443 static int 1444 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, const struct flowi *fl, 1445 struct xfrm_state **xfrm, unsigned short family) 1446 { 1447 struct net *net = xp_net(policy); 1448 int nx; 1449 int i, error; 1450 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family); 1451 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family); 1452 xfrm_address_t tmp; 1453 1454 for (nx = 0, i = 0; i < policy->xfrm_nr; i++) { 1455 struct xfrm_state *x; 1456 xfrm_address_t *remote = daddr; 1457 xfrm_address_t *local = saddr; 1458 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i]; 1459 1460 if (tmpl->mode == XFRM_MODE_TUNNEL || 1461 tmpl->mode == XFRM_MODE_BEET) { 1462 remote = &tmpl->id.daddr; 1463 local = &tmpl->saddr; 1464 if (xfrm_addr_any(local, tmpl->encap_family)) { 1465 error = xfrm_get_saddr(net, fl->flowi_oif, 1466 &tmp, remote, 1467 tmpl->encap_family); 1468 if (error) 1469 goto fail; 1470 local = &tmp; 1471 } 1472 } 1473 1474 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family); 1475 1476 if (x && x->km.state == XFRM_STATE_VALID) { 1477 xfrm[nx++] = x; 1478 daddr = remote; 1479 saddr = local; 1480 continue; 1481 } 1482 if (x) { 1483 error = (x->km.state == XFRM_STATE_ERROR ? 1484 -EINVAL : -EAGAIN); 1485 xfrm_state_put(x); 1486 } else if (error == -ESRCH) { 1487 error = -EAGAIN; 1488 } 1489 1490 if (!tmpl->optional) 1491 goto fail; 1492 } 1493 return nx; 1494 1495 fail: 1496 for (nx--; nx >= 0; nx--) 1497 xfrm_state_put(xfrm[nx]); 1498 return error; 1499 } 1500 1501 static int 1502 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, const struct flowi *fl, 1503 struct xfrm_state **xfrm, unsigned short family) 1504 { 1505 struct xfrm_state *tp[XFRM_MAX_DEPTH]; 1506 struct xfrm_state **tpp = (npols > 1) ? tp : xfrm; 1507 int cnx = 0; 1508 int error; 1509 int ret; 1510 int i; 1511 1512 for (i = 0; i < npols; i++) { 1513 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) { 1514 error = -ENOBUFS; 1515 goto fail; 1516 } 1517 1518 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family); 1519 if (ret < 0) { 1520 error = ret; 1521 goto fail; 1522 } else 1523 cnx += ret; 1524 } 1525 1526 /* found states are sorted for outbound processing */ 1527 if (npols > 1) 1528 xfrm_state_sort(xfrm, tpp, cnx, family); 1529 1530 return cnx; 1531 1532 fail: 1533 for (cnx--; cnx >= 0; cnx--) 1534 xfrm_state_put(tpp[cnx]); 1535 return error; 1536 1537 } 1538 1539 static int xfrm_get_tos(const struct flowi *fl, int family) 1540 { 1541 const struct xfrm_policy_afinfo *afinfo; 1542 int tos = 0; 1543 1544 afinfo = xfrm_policy_get_afinfo(family); 1545 tos = afinfo ? afinfo->get_tos(fl) : 0; 1546 1547 rcu_read_unlock(); 1548 1549 return tos; 1550 } 1551 1552 static struct flow_cache_object *xfrm_bundle_flo_get(struct flow_cache_object *flo) 1553 { 1554 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo); 1555 struct dst_entry *dst = &xdst->u.dst; 1556 1557 if (xdst->route == NULL) { 1558 /* Dummy bundle - if it has xfrms we were not 1559 * able to build bundle as template resolution failed. 1560 * It means we need to try again resolving. */ 1561 if (xdst->num_xfrms > 0) 1562 return NULL; 1563 } else if (dst->flags & DST_XFRM_QUEUE) { 1564 return NULL; 1565 } else { 1566 /* Real bundle */ 1567 if (stale_bundle(dst)) 1568 return NULL; 1569 } 1570 1571 dst_hold(dst); 1572 return flo; 1573 } 1574 1575 static int xfrm_bundle_flo_check(struct flow_cache_object *flo) 1576 { 1577 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo); 1578 struct dst_entry *dst = &xdst->u.dst; 1579 1580 if (!xdst->route) 1581 return 0; 1582 if (stale_bundle(dst)) 1583 return 0; 1584 1585 return 1; 1586 } 1587 1588 static void xfrm_bundle_flo_delete(struct flow_cache_object *flo) 1589 { 1590 struct xfrm_dst *xdst = container_of(flo, struct xfrm_dst, flo); 1591 struct dst_entry *dst = &xdst->u.dst; 1592 1593 dst_free(dst); 1594 } 1595 1596 static const struct flow_cache_ops xfrm_bundle_fc_ops = { 1597 .get = xfrm_bundle_flo_get, 1598 .check = xfrm_bundle_flo_check, 1599 .delete = xfrm_bundle_flo_delete, 1600 }; 1601 1602 static inline struct xfrm_dst *xfrm_alloc_dst(struct net *net, int family) 1603 { 1604 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 1605 struct dst_ops *dst_ops; 1606 struct xfrm_dst *xdst; 1607 1608 if (!afinfo) 1609 return ERR_PTR(-EINVAL); 1610 1611 switch (family) { 1612 case AF_INET: 1613 dst_ops = &net->xfrm.xfrm4_dst_ops; 1614 break; 1615 #if IS_ENABLED(CONFIG_IPV6) 1616 case AF_INET6: 1617 dst_ops = &net->xfrm.xfrm6_dst_ops; 1618 break; 1619 #endif 1620 default: 1621 BUG(); 1622 } 1623 xdst = dst_alloc(dst_ops, NULL, 0, DST_OBSOLETE_NONE, 0); 1624 1625 if (likely(xdst)) { 1626 struct dst_entry *dst = &xdst->u.dst; 1627 1628 memset(dst + 1, 0, sizeof(*xdst) - sizeof(*dst)); 1629 xdst->flo.ops = &xfrm_bundle_fc_ops; 1630 } else 1631 xdst = ERR_PTR(-ENOBUFS); 1632 1633 rcu_read_unlock(); 1634 1635 return xdst; 1636 } 1637 1638 static inline int xfrm_init_path(struct xfrm_dst *path, struct dst_entry *dst, 1639 int nfheader_len) 1640 { 1641 const struct xfrm_policy_afinfo *afinfo = 1642 xfrm_policy_get_afinfo(dst->ops->family); 1643 int err; 1644 1645 if (!afinfo) 1646 return -EINVAL; 1647 1648 err = afinfo->init_path(path, dst, nfheader_len); 1649 1650 rcu_read_unlock(); 1651 1652 return err; 1653 } 1654 1655 static inline int xfrm_fill_dst(struct xfrm_dst *xdst, struct net_device *dev, 1656 const struct flowi *fl) 1657 { 1658 const struct xfrm_policy_afinfo *afinfo = 1659 xfrm_policy_get_afinfo(xdst->u.dst.ops->family); 1660 int err; 1661 1662 if (!afinfo) 1663 return -EINVAL; 1664 1665 err = afinfo->fill_dst(xdst, dev, fl); 1666 1667 rcu_read_unlock(); 1668 1669 return err; 1670 } 1671 1672 1673 /* Allocate chain of dst_entry's, attach known xfrm's, calculate 1674 * all the metrics... Shortly, bundle a bundle. 1675 */ 1676 1677 static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy, 1678 struct xfrm_state **xfrm, int nx, 1679 const struct flowi *fl, 1680 struct dst_entry *dst) 1681 { 1682 struct net *net = xp_net(policy); 1683 unsigned long now = jiffies; 1684 struct net_device *dev; 1685 struct xfrm_mode *inner_mode; 1686 struct dst_entry *dst_prev = NULL; 1687 struct dst_entry *dst0 = NULL; 1688 int i = 0; 1689 int err; 1690 int header_len = 0; 1691 int nfheader_len = 0; 1692 int trailer_len = 0; 1693 int tos; 1694 int family = policy->selector.family; 1695 xfrm_address_t saddr, daddr; 1696 1697 xfrm_flowi_addr_get(fl, &saddr, &daddr, family); 1698 1699 tos = xfrm_get_tos(fl, family); 1700 1701 dst_hold(dst); 1702 1703 for (; i < nx; i++) { 1704 struct xfrm_dst *xdst = xfrm_alloc_dst(net, family); 1705 struct dst_entry *dst1 = &xdst->u.dst; 1706 1707 err = PTR_ERR(xdst); 1708 if (IS_ERR(xdst)) { 1709 dst_release(dst); 1710 goto put_states; 1711 } 1712 1713 if (xfrm[i]->sel.family == AF_UNSPEC) { 1714 inner_mode = xfrm_ip2inner_mode(xfrm[i], 1715 xfrm_af2proto(family)); 1716 if (!inner_mode) { 1717 err = -EAFNOSUPPORT; 1718 dst_release(dst); 1719 goto put_states; 1720 } 1721 } else 1722 inner_mode = xfrm[i]->inner_mode; 1723 1724 if (!dst_prev) 1725 dst0 = dst1; 1726 else { 1727 dst_prev->child = dst_clone(dst1); 1728 dst1->flags |= DST_NOHASH; 1729 } 1730 1731 xdst->route = dst; 1732 dst_copy_metrics(dst1, dst); 1733 1734 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) { 1735 family = xfrm[i]->props.family; 1736 dst = xfrm_dst_lookup(xfrm[i], tos, fl->flowi_oif, 1737 &saddr, &daddr, family); 1738 err = PTR_ERR(dst); 1739 if (IS_ERR(dst)) 1740 goto put_states; 1741 } else 1742 dst_hold(dst); 1743 1744 dst1->xfrm = xfrm[i]; 1745 xdst->xfrm_genid = xfrm[i]->genid; 1746 1747 dst1->obsolete = DST_OBSOLETE_FORCE_CHK; 1748 dst1->flags |= DST_HOST; 1749 dst1->lastuse = now; 1750 1751 dst1->input = dst_discard; 1752 dst1->output = inner_mode->afinfo->output; 1753 1754 dst1->next = dst_prev; 1755 dst_prev = dst1; 1756 1757 header_len += xfrm[i]->props.header_len; 1758 if (xfrm[i]->type->flags & XFRM_TYPE_NON_FRAGMENT) 1759 nfheader_len += xfrm[i]->props.header_len; 1760 trailer_len += xfrm[i]->props.trailer_len; 1761 } 1762 1763 dst_prev->child = dst; 1764 dst0->path = dst; 1765 1766 err = -ENODEV; 1767 dev = dst->dev; 1768 if (!dev) 1769 goto free_dst; 1770 1771 xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len); 1772 xfrm_init_pmtu(dst_prev); 1773 1774 for (dst_prev = dst0; dst_prev != dst; dst_prev = dst_prev->child) { 1775 struct xfrm_dst *xdst = (struct xfrm_dst *)dst_prev; 1776 1777 err = xfrm_fill_dst(xdst, dev, fl); 1778 if (err) 1779 goto free_dst; 1780 1781 dst_prev->header_len = header_len; 1782 dst_prev->trailer_len = trailer_len; 1783 header_len -= xdst->u.dst.xfrm->props.header_len; 1784 trailer_len -= xdst->u.dst.xfrm->props.trailer_len; 1785 } 1786 1787 out: 1788 return dst0; 1789 1790 put_states: 1791 for (; i < nx; i++) 1792 xfrm_state_put(xfrm[i]); 1793 free_dst: 1794 if (dst0) 1795 dst_free(dst0); 1796 dst0 = ERR_PTR(err); 1797 goto out; 1798 } 1799 1800 #ifdef CONFIG_XFRM_SUB_POLICY 1801 static int xfrm_dst_alloc_copy(void **target, const void *src, int size) 1802 { 1803 if (!*target) { 1804 *target = kmalloc(size, GFP_ATOMIC); 1805 if (!*target) 1806 return -ENOMEM; 1807 } 1808 1809 memcpy(*target, src, size); 1810 return 0; 1811 } 1812 #endif 1813 1814 static int xfrm_dst_update_parent(struct dst_entry *dst, 1815 const struct xfrm_selector *sel) 1816 { 1817 #ifdef CONFIG_XFRM_SUB_POLICY 1818 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 1819 return xfrm_dst_alloc_copy((void **)&(xdst->partner), 1820 sel, sizeof(*sel)); 1821 #else 1822 return 0; 1823 #endif 1824 } 1825 1826 static int xfrm_dst_update_origin(struct dst_entry *dst, 1827 const struct flowi *fl) 1828 { 1829 #ifdef CONFIG_XFRM_SUB_POLICY 1830 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 1831 return xfrm_dst_alloc_copy((void **)&(xdst->origin), fl, sizeof(*fl)); 1832 #else 1833 return 0; 1834 #endif 1835 } 1836 1837 static int xfrm_expand_policies(const struct flowi *fl, u16 family, 1838 struct xfrm_policy **pols, 1839 int *num_pols, int *num_xfrms) 1840 { 1841 int i; 1842 1843 if (*num_pols == 0 || !pols[0]) { 1844 *num_pols = 0; 1845 *num_xfrms = 0; 1846 return 0; 1847 } 1848 if (IS_ERR(pols[0])) 1849 return PTR_ERR(pols[0]); 1850 1851 *num_xfrms = pols[0]->xfrm_nr; 1852 1853 #ifdef CONFIG_XFRM_SUB_POLICY 1854 if (pols[0] && pols[0]->action == XFRM_POLICY_ALLOW && 1855 pols[0]->type != XFRM_POLICY_TYPE_MAIN) { 1856 pols[1] = xfrm_policy_lookup_bytype(xp_net(pols[0]), 1857 XFRM_POLICY_TYPE_MAIN, 1858 fl, family, 1859 XFRM_POLICY_OUT); 1860 if (pols[1]) { 1861 if (IS_ERR(pols[1])) { 1862 xfrm_pols_put(pols, *num_pols); 1863 return PTR_ERR(pols[1]); 1864 } 1865 (*num_pols)++; 1866 (*num_xfrms) += pols[1]->xfrm_nr; 1867 } 1868 } 1869 #endif 1870 for (i = 0; i < *num_pols; i++) { 1871 if (pols[i]->action != XFRM_POLICY_ALLOW) { 1872 *num_xfrms = -1; 1873 break; 1874 } 1875 } 1876 1877 return 0; 1878 1879 } 1880 1881 static struct xfrm_dst * 1882 xfrm_resolve_and_create_bundle(struct xfrm_policy **pols, int num_pols, 1883 const struct flowi *fl, u16 family, 1884 struct dst_entry *dst_orig) 1885 { 1886 struct net *net = xp_net(pols[0]); 1887 struct xfrm_state *xfrm[XFRM_MAX_DEPTH]; 1888 struct dst_entry *dst; 1889 struct xfrm_dst *xdst; 1890 int err; 1891 1892 /* Try to instantiate a bundle */ 1893 err = xfrm_tmpl_resolve(pols, num_pols, fl, xfrm, family); 1894 if (err <= 0) { 1895 if (err != 0 && err != -EAGAIN) 1896 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR); 1897 return ERR_PTR(err); 1898 } 1899 1900 dst = xfrm_bundle_create(pols[0], xfrm, err, fl, dst_orig); 1901 if (IS_ERR(dst)) { 1902 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR); 1903 return ERR_CAST(dst); 1904 } 1905 1906 xdst = (struct xfrm_dst *)dst; 1907 xdst->num_xfrms = err; 1908 if (num_pols > 1) 1909 err = xfrm_dst_update_parent(dst, &pols[1]->selector); 1910 else 1911 err = xfrm_dst_update_origin(dst, fl); 1912 if (unlikely(err)) { 1913 dst_free(dst); 1914 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR); 1915 return ERR_PTR(err); 1916 } 1917 1918 xdst->num_pols = num_pols; 1919 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols); 1920 xdst->policy_genid = atomic_read(&pols[0]->genid); 1921 1922 return xdst; 1923 } 1924 1925 static void xfrm_policy_queue_process(unsigned long arg) 1926 { 1927 struct sk_buff *skb; 1928 struct sock *sk; 1929 struct dst_entry *dst; 1930 struct xfrm_policy *pol = (struct xfrm_policy *)arg; 1931 struct net *net = xp_net(pol); 1932 struct xfrm_policy_queue *pq = &pol->polq; 1933 struct flowi fl; 1934 struct sk_buff_head list; 1935 1936 spin_lock(&pq->hold_queue.lock); 1937 skb = skb_peek(&pq->hold_queue); 1938 if (!skb) { 1939 spin_unlock(&pq->hold_queue.lock); 1940 goto out; 1941 } 1942 dst = skb_dst(skb); 1943 sk = skb->sk; 1944 xfrm_decode_session(skb, &fl, dst->ops->family); 1945 spin_unlock(&pq->hold_queue.lock); 1946 1947 dst_hold(dst->path); 1948 dst = xfrm_lookup(net, dst->path, &fl, sk, 0); 1949 if (IS_ERR(dst)) 1950 goto purge_queue; 1951 1952 if (dst->flags & DST_XFRM_QUEUE) { 1953 dst_release(dst); 1954 1955 if (pq->timeout >= XFRM_QUEUE_TMO_MAX) 1956 goto purge_queue; 1957 1958 pq->timeout = pq->timeout << 1; 1959 if (!mod_timer(&pq->hold_timer, jiffies + pq->timeout)) 1960 xfrm_pol_hold(pol); 1961 goto out; 1962 } 1963 1964 dst_release(dst); 1965 1966 __skb_queue_head_init(&list); 1967 1968 spin_lock(&pq->hold_queue.lock); 1969 pq->timeout = 0; 1970 skb_queue_splice_init(&pq->hold_queue, &list); 1971 spin_unlock(&pq->hold_queue.lock); 1972 1973 while (!skb_queue_empty(&list)) { 1974 skb = __skb_dequeue(&list); 1975 1976 xfrm_decode_session(skb, &fl, skb_dst(skb)->ops->family); 1977 dst_hold(skb_dst(skb)->path); 1978 dst = xfrm_lookup(net, skb_dst(skb)->path, &fl, skb->sk, 0); 1979 if (IS_ERR(dst)) { 1980 kfree_skb(skb); 1981 continue; 1982 } 1983 1984 nf_reset(skb); 1985 skb_dst_drop(skb); 1986 skb_dst_set(skb, dst); 1987 1988 dst_output(net, skb->sk, skb); 1989 } 1990 1991 out: 1992 xfrm_pol_put(pol); 1993 return; 1994 1995 purge_queue: 1996 pq->timeout = 0; 1997 skb_queue_purge(&pq->hold_queue); 1998 xfrm_pol_put(pol); 1999 } 2000 2001 static int xdst_queue_output(struct net *net, struct sock *sk, struct sk_buff *skb) 2002 { 2003 unsigned long sched_next; 2004 struct dst_entry *dst = skb_dst(skb); 2005 struct xfrm_dst *xdst = (struct xfrm_dst *) dst; 2006 struct xfrm_policy *pol = xdst->pols[0]; 2007 struct xfrm_policy_queue *pq = &pol->polq; 2008 2009 if (unlikely(skb_fclone_busy(sk, skb))) { 2010 kfree_skb(skb); 2011 return 0; 2012 } 2013 2014 if (pq->hold_queue.qlen > XFRM_MAX_QUEUE_LEN) { 2015 kfree_skb(skb); 2016 return -EAGAIN; 2017 } 2018 2019 skb_dst_force(skb); 2020 2021 spin_lock_bh(&pq->hold_queue.lock); 2022 2023 if (!pq->timeout) 2024 pq->timeout = XFRM_QUEUE_TMO_MIN; 2025 2026 sched_next = jiffies + pq->timeout; 2027 2028 if (del_timer(&pq->hold_timer)) { 2029 if (time_before(pq->hold_timer.expires, sched_next)) 2030 sched_next = pq->hold_timer.expires; 2031 xfrm_pol_put(pol); 2032 } 2033 2034 __skb_queue_tail(&pq->hold_queue, skb); 2035 if (!mod_timer(&pq->hold_timer, sched_next)) 2036 xfrm_pol_hold(pol); 2037 2038 spin_unlock_bh(&pq->hold_queue.lock); 2039 2040 return 0; 2041 } 2042 2043 static struct xfrm_dst *xfrm_create_dummy_bundle(struct net *net, 2044 struct xfrm_flo *xflo, 2045 const struct flowi *fl, 2046 int num_xfrms, 2047 u16 family) 2048 { 2049 int err; 2050 struct net_device *dev; 2051 struct dst_entry *dst; 2052 struct dst_entry *dst1; 2053 struct xfrm_dst *xdst; 2054 2055 xdst = xfrm_alloc_dst(net, family); 2056 if (IS_ERR(xdst)) 2057 return xdst; 2058 2059 if (!(xflo->flags & XFRM_LOOKUP_QUEUE) || 2060 net->xfrm.sysctl_larval_drop || 2061 num_xfrms <= 0) 2062 return xdst; 2063 2064 dst = xflo->dst_orig; 2065 dst1 = &xdst->u.dst; 2066 dst_hold(dst); 2067 xdst->route = dst; 2068 2069 dst_copy_metrics(dst1, dst); 2070 2071 dst1->obsolete = DST_OBSOLETE_FORCE_CHK; 2072 dst1->flags |= DST_HOST | DST_XFRM_QUEUE; 2073 dst1->lastuse = jiffies; 2074 2075 dst1->input = dst_discard; 2076 dst1->output = xdst_queue_output; 2077 2078 dst_hold(dst); 2079 dst1->child = dst; 2080 dst1->path = dst; 2081 2082 xfrm_init_path((struct xfrm_dst *)dst1, dst, 0); 2083 2084 err = -ENODEV; 2085 dev = dst->dev; 2086 if (!dev) 2087 goto free_dst; 2088 2089 err = xfrm_fill_dst(xdst, dev, fl); 2090 if (err) 2091 goto free_dst; 2092 2093 out: 2094 return xdst; 2095 2096 free_dst: 2097 dst_release(dst1); 2098 xdst = ERR_PTR(err); 2099 goto out; 2100 } 2101 2102 static struct flow_cache_object * 2103 xfrm_bundle_lookup(struct net *net, const struct flowi *fl, u16 family, u8 dir, 2104 struct flow_cache_object *oldflo, void *ctx) 2105 { 2106 struct xfrm_flo *xflo = (struct xfrm_flo *)ctx; 2107 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 2108 struct xfrm_dst *xdst, *new_xdst; 2109 int num_pols = 0, num_xfrms = 0, i, err, pol_dead; 2110 2111 /* Check if the policies from old bundle are usable */ 2112 xdst = NULL; 2113 if (oldflo) { 2114 xdst = container_of(oldflo, struct xfrm_dst, flo); 2115 num_pols = xdst->num_pols; 2116 num_xfrms = xdst->num_xfrms; 2117 pol_dead = 0; 2118 for (i = 0; i < num_pols; i++) { 2119 pols[i] = xdst->pols[i]; 2120 pol_dead |= pols[i]->walk.dead; 2121 } 2122 if (pol_dead) { 2123 dst_free(&xdst->u.dst); 2124 xdst = NULL; 2125 num_pols = 0; 2126 num_xfrms = 0; 2127 oldflo = NULL; 2128 } 2129 } 2130 2131 /* Resolve policies to use if we couldn't get them from 2132 * previous cache entry */ 2133 if (xdst == NULL) { 2134 num_pols = 1; 2135 pols[0] = __xfrm_policy_lookup(net, fl, family, 2136 flow_to_policy_dir(dir)); 2137 err = xfrm_expand_policies(fl, family, pols, 2138 &num_pols, &num_xfrms); 2139 if (err < 0) 2140 goto inc_error; 2141 if (num_pols == 0) 2142 return NULL; 2143 if (num_xfrms <= 0) 2144 goto make_dummy_bundle; 2145 } 2146 2147 new_xdst = xfrm_resolve_and_create_bundle(pols, num_pols, fl, family, 2148 xflo->dst_orig); 2149 if (IS_ERR(new_xdst)) { 2150 err = PTR_ERR(new_xdst); 2151 if (err != -EAGAIN) 2152 goto error; 2153 if (oldflo == NULL) 2154 goto make_dummy_bundle; 2155 dst_hold(&xdst->u.dst); 2156 return oldflo; 2157 } else if (new_xdst == NULL) { 2158 num_xfrms = 0; 2159 if (oldflo == NULL) 2160 goto make_dummy_bundle; 2161 xdst->num_xfrms = 0; 2162 dst_hold(&xdst->u.dst); 2163 return oldflo; 2164 } 2165 2166 /* Kill the previous bundle */ 2167 if (xdst) { 2168 /* The policies were stolen for newly generated bundle */ 2169 xdst->num_pols = 0; 2170 dst_free(&xdst->u.dst); 2171 } 2172 2173 /* Flow cache does not have reference, it dst_free()'s, 2174 * but we do need to return one reference for original caller */ 2175 dst_hold(&new_xdst->u.dst); 2176 return &new_xdst->flo; 2177 2178 make_dummy_bundle: 2179 /* We found policies, but there's no bundles to instantiate: 2180 * either because the policy blocks, has no transformations or 2181 * we could not build template (no xfrm_states).*/ 2182 xdst = xfrm_create_dummy_bundle(net, xflo, fl, num_xfrms, family); 2183 if (IS_ERR(xdst)) { 2184 xfrm_pols_put(pols, num_pols); 2185 return ERR_CAST(xdst); 2186 } 2187 xdst->num_pols = num_pols; 2188 xdst->num_xfrms = num_xfrms; 2189 memcpy(xdst->pols, pols, sizeof(struct xfrm_policy *) * num_pols); 2190 2191 dst_hold(&xdst->u.dst); 2192 return &xdst->flo; 2193 2194 inc_error: 2195 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR); 2196 error: 2197 if (xdst != NULL) 2198 dst_free(&xdst->u.dst); 2199 else 2200 xfrm_pols_put(pols, num_pols); 2201 return ERR_PTR(err); 2202 } 2203 2204 static struct dst_entry *make_blackhole(struct net *net, u16 family, 2205 struct dst_entry *dst_orig) 2206 { 2207 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 2208 struct dst_entry *ret; 2209 2210 if (!afinfo) { 2211 dst_release(dst_orig); 2212 return ERR_PTR(-EINVAL); 2213 } else { 2214 ret = afinfo->blackhole_route(net, dst_orig); 2215 } 2216 rcu_read_unlock(); 2217 2218 return ret; 2219 } 2220 2221 /* Main function: finds/creates a bundle for given flow. 2222 * 2223 * At the moment we eat a raw IP route. Mostly to speed up lookups 2224 * on interfaces with disabled IPsec. 2225 */ 2226 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig, 2227 const struct flowi *fl, 2228 const struct sock *sk, int flags) 2229 { 2230 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 2231 struct flow_cache_object *flo; 2232 struct xfrm_dst *xdst; 2233 struct dst_entry *dst, *route; 2234 u16 family = dst_orig->ops->family; 2235 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT); 2236 int i, err, num_pols, num_xfrms = 0, drop_pols = 0; 2237 2238 dst = NULL; 2239 xdst = NULL; 2240 route = NULL; 2241 2242 sk = sk_const_to_full_sk(sk); 2243 if (sk && sk->sk_policy[XFRM_POLICY_OUT]) { 2244 num_pols = 1; 2245 pols[0] = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, family); 2246 err = xfrm_expand_policies(fl, family, pols, 2247 &num_pols, &num_xfrms); 2248 if (err < 0) 2249 goto dropdst; 2250 2251 if (num_pols) { 2252 if (num_xfrms <= 0) { 2253 drop_pols = num_pols; 2254 goto no_transform; 2255 } 2256 2257 xdst = xfrm_resolve_and_create_bundle( 2258 pols, num_pols, fl, 2259 family, dst_orig); 2260 if (IS_ERR(xdst)) { 2261 xfrm_pols_put(pols, num_pols); 2262 err = PTR_ERR(xdst); 2263 goto dropdst; 2264 } else if (xdst == NULL) { 2265 num_xfrms = 0; 2266 drop_pols = num_pols; 2267 goto no_transform; 2268 } 2269 2270 dst_hold(&xdst->u.dst); 2271 xdst->u.dst.flags |= DST_NOCACHE; 2272 route = xdst->route; 2273 } 2274 } 2275 2276 if (xdst == NULL) { 2277 struct xfrm_flo xflo; 2278 2279 xflo.dst_orig = dst_orig; 2280 xflo.flags = flags; 2281 2282 /* To accelerate a bit... */ 2283 if ((dst_orig->flags & DST_NOXFRM) || 2284 !net->xfrm.policy_count[XFRM_POLICY_OUT]) 2285 goto nopol; 2286 2287 flo = flow_cache_lookup(net, fl, family, dir, 2288 xfrm_bundle_lookup, &xflo); 2289 if (flo == NULL) 2290 goto nopol; 2291 if (IS_ERR(flo)) { 2292 err = PTR_ERR(flo); 2293 goto dropdst; 2294 } 2295 xdst = container_of(flo, struct xfrm_dst, flo); 2296 2297 num_pols = xdst->num_pols; 2298 num_xfrms = xdst->num_xfrms; 2299 memcpy(pols, xdst->pols, sizeof(struct xfrm_policy *) * num_pols); 2300 route = xdst->route; 2301 } 2302 2303 dst = &xdst->u.dst; 2304 if (route == NULL && num_xfrms > 0) { 2305 /* The only case when xfrm_bundle_lookup() returns a 2306 * bundle with null route, is when the template could 2307 * not be resolved. It means policies are there, but 2308 * bundle could not be created, since we don't yet 2309 * have the xfrm_state's. We need to wait for KM to 2310 * negotiate new SA's or bail out with error.*/ 2311 if (net->xfrm.sysctl_larval_drop) { 2312 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES); 2313 err = -EREMOTE; 2314 goto error; 2315 } 2316 2317 err = -EAGAIN; 2318 2319 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES); 2320 goto error; 2321 } 2322 2323 no_transform: 2324 if (num_pols == 0) 2325 goto nopol; 2326 2327 if ((flags & XFRM_LOOKUP_ICMP) && 2328 !(pols[0]->flags & XFRM_POLICY_ICMP)) { 2329 err = -ENOENT; 2330 goto error; 2331 } 2332 2333 for (i = 0; i < num_pols; i++) 2334 pols[i]->curlft.use_time = get_seconds(); 2335 2336 if (num_xfrms < 0) { 2337 /* Prohibit the flow */ 2338 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK); 2339 err = -EPERM; 2340 goto error; 2341 } else if (num_xfrms > 0) { 2342 /* Flow transformed */ 2343 dst_release(dst_orig); 2344 } else { 2345 /* Flow passes untransformed */ 2346 dst_release(dst); 2347 dst = dst_orig; 2348 } 2349 ok: 2350 xfrm_pols_put(pols, drop_pols); 2351 if (dst && dst->xfrm && 2352 dst->xfrm->props.mode == XFRM_MODE_TUNNEL) 2353 dst->flags |= DST_XFRM_TUNNEL; 2354 return dst; 2355 2356 nopol: 2357 if (!(flags & XFRM_LOOKUP_ICMP)) { 2358 dst = dst_orig; 2359 goto ok; 2360 } 2361 err = -ENOENT; 2362 error: 2363 dst_release(dst); 2364 dropdst: 2365 if (!(flags & XFRM_LOOKUP_KEEP_DST_REF)) 2366 dst_release(dst_orig); 2367 xfrm_pols_put(pols, drop_pols); 2368 return ERR_PTR(err); 2369 } 2370 EXPORT_SYMBOL(xfrm_lookup); 2371 2372 /* Callers of xfrm_lookup_route() must ensure a call to dst_output(). 2373 * Otherwise we may send out blackholed packets. 2374 */ 2375 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig, 2376 const struct flowi *fl, 2377 const struct sock *sk, int flags) 2378 { 2379 struct dst_entry *dst = xfrm_lookup(net, dst_orig, fl, sk, 2380 flags | XFRM_LOOKUP_QUEUE | 2381 XFRM_LOOKUP_KEEP_DST_REF); 2382 2383 if (IS_ERR(dst) && PTR_ERR(dst) == -EREMOTE) 2384 return make_blackhole(net, dst_orig->ops->family, dst_orig); 2385 2386 return dst; 2387 } 2388 EXPORT_SYMBOL(xfrm_lookup_route); 2389 2390 static inline int 2391 xfrm_secpath_reject(int idx, struct sk_buff *skb, const struct flowi *fl) 2392 { 2393 struct xfrm_state *x; 2394 2395 if (!skb->sp || idx < 0 || idx >= skb->sp->len) 2396 return 0; 2397 x = skb->sp->xvec[idx]; 2398 if (!x->type->reject) 2399 return 0; 2400 return x->type->reject(x, skb, fl); 2401 } 2402 2403 /* When skb is transformed back to its "native" form, we have to 2404 * check policy restrictions. At the moment we make this in maximally 2405 * stupid way. Shame on me. :-) Of course, connected sockets must 2406 * have policy cached at them. 2407 */ 2408 2409 static inline int 2410 xfrm_state_ok(const struct xfrm_tmpl *tmpl, const struct xfrm_state *x, 2411 unsigned short family) 2412 { 2413 if (xfrm_state_kern(x)) 2414 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, tmpl->encap_family); 2415 return x->id.proto == tmpl->id.proto && 2416 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) && 2417 (x->props.reqid == tmpl->reqid || !tmpl->reqid) && 2418 x->props.mode == tmpl->mode && 2419 (tmpl->allalgs || (tmpl->aalgos & (1<<x->props.aalgo)) || 2420 !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) && 2421 !(x->props.mode != XFRM_MODE_TRANSPORT && 2422 xfrm_state_addr_cmp(tmpl, x, family)); 2423 } 2424 2425 /* 2426 * 0 or more than 0 is returned when validation is succeeded (either bypass 2427 * because of optional transport mode, or next index of the mathced secpath 2428 * state with the template. 2429 * -1 is returned when no matching template is found. 2430 * Otherwise "-2 - errored_index" is returned. 2431 */ 2432 static inline int 2433 xfrm_policy_ok(const struct xfrm_tmpl *tmpl, const struct sec_path *sp, int start, 2434 unsigned short family) 2435 { 2436 int idx = start; 2437 2438 if (tmpl->optional) { 2439 if (tmpl->mode == XFRM_MODE_TRANSPORT) 2440 return start; 2441 } else 2442 start = -1; 2443 for (; idx < sp->len; idx++) { 2444 if (xfrm_state_ok(tmpl, sp->xvec[idx], family)) 2445 return ++idx; 2446 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) { 2447 if (start == -1) 2448 start = -2-idx; 2449 break; 2450 } 2451 } 2452 return start; 2453 } 2454 2455 int __xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, 2456 unsigned int family, int reverse) 2457 { 2458 const struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 2459 int err; 2460 2461 if (unlikely(afinfo == NULL)) 2462 return -EAFNOSUPPORT; 2463 2464 afinfo->decode_session(skb, fl, reverse); 2465 err = security_xfrm_decode_session(skb, &fl->flowi_secid); 2466 rcu_read_unlock(); 2467 return err; 2468 } 2469 EXPORT_SYMBOL(__xfrm_decode_session); 2470 2471 static inline int secpath_has_nontransport(const struct sec_path *sp, int k, int *idxp) 2472 { 2473 for (; k < sp->len; k++) { 2474 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) { 2475 *idxp = k; 2476 return 1; 2477 } 2478 } 2479 2480 return 0; 2481 } 2482 2483 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 2484 unsigned short family) 2485 { 2486 struct net *net = dev_net(skb->dev); 2487 struct xfrm_policy *pol; 2488 struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX]; 2489 int npols = 0; 2490 int xfrm_nr; 2491 int pi; 2492 int reverse; 2493 struct flowi fl; 2494 u8 fl_dir; 2495 int xerr_idx = -1; 2496 2497 reverse = dir & ~XFRM_POLICY_MASK; 2498 dir &= XFRM_POLICY_MASK; 2499 fl_dir = policy_to_flow_dir(dir); 2500 2501 if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) { 2502 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR); 2503 return 0; 2504 } 2505 2506 nf_nat_decode_session(skb, &fl, family); 2507 2508 /* First, check used SA against their selectors. */ 2509 if (skb->sp) { 2510 int i; 2511 2512 for (i = skb->sp->len-1; i >= 0; i--) { 2513 struct xfrm_state *x = skb->sp->xvec[i]; 2514 if (!xfrm_selector_match(&x->sel, &fl, family)) { 2515 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH); 2516 return 0; 2517 } 2518 } 2519 } 2520 2521 pol = NULL; 2522 sk = sk_to_full_sk(sk); 2523 if (sk && sk->sk_policy[dir]) { 2524 pol = xfrm_sk_policy_lookup(sk, dir, &fl, family); 2525 if (IS_ERR(pol)) { 2526 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 2527 return 0; 2528 } 2529 } 2530 2531 if (!pol) { 2532 struct flow_cache_object *flo; 2533 2534 flo = flow_cache_lookup(net, &fl, family, fl_dir, 2535 xfrm_policy_lookup, NULL); 2536 if (IS_ERR_OR_NULL(flo)) 2537 pol = ERR_CAST(flo); 2538 else 2539 pol = container_of(flo, struct xfrm_policy, flo); 2540 } 2541 2542 if (IS_ERR(pol)) { 2543 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 2544 return 0; 2545 } 2546 2547 if (!pol) { 2548 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) { 2549 xfrm_secpath_reject(xerr_idx, skb, &fl); 2550 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS); 2551 return 0; 2552 } 2553 return 1; 2554 } 2555 2556 pol->curlft.use_time = get_seconds(); 2557 2558 pols[0] = pol; 2559 npols++; 2560 #ifdef CONFIG_XFRM_SUB_POLICY 2561 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) { 2562 pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, 2563 &fl, family, 2564 XFRM_POLICY_IN); 2565 if (pols[1]) { 2566 if (IS_ERR(pols[1])) { 2567 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR); 2568 return 0; 2569 } 2570 pols[1]->curlft.use_time = get_seconds(); 2571 npols++; 2572 } 2573 } 2574 #endif 2575 2576 if (pol->action == XFRM_POLICY_ALLOW) { 2577 struct sec_path *sp; 2578 static struct sec_path dummy; 2579 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH]; 2580 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH]; 2581 struct xfrm_tmpl **tpp = tp; 2582 int ti = 0; 2583 int i, k; 2584 2585 if ((sp = skb->sp) == NULL) 2586 sp = &dummy; 2587 2588 for (pi = 0; pi < npols; pi++) { 2589 if (pols[pi] != pol && 2590 pols[pi]->action != XFRM_POLICY_ALLOW) { 2591 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK); 2592 goto reject; 2593 } 2594 if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) { 2595 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR); 2596 goto reject_error; 2597 } 2598 for (i = 0; i < pols[pi]->xfrm_nr; i++) 2599 tpp[ti++] = &pols[pi]->xfrm_vec[i]; 2600 } 2601 xfrm_nr = ti; 2602 if (npols > 1) { 2603 xfrm_tmpl_sort(stp, tpp, xfrm_nr, family, net); 2604 tpp = stp; 2605 } 2606 2607 /* For each tunnel xfrm, find the first matching tmpl. 2608 * For each tmpl before that, find corresponding xfrm. 2609 * Order is _important_. Later we will implement 2610 * some barriers, but at the moment barriers 2611 * are implied between each two transformations. 2612 */ 2613 for (i = xfrm_nr-1, k = 0; i >= 0; i--) { 2614 k = xfrm_policy_ok(tpp[i], sp, k, family); 2615 if (k < 0) { 2616 if (k < -1) 2617 /* "-2 - errored_index" returned */ 2618 xerr_idx = -(2+k); 2619 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH); 2620 goto reject; 2621 } 2622 } 2623 2624 if (secpath_has_nontransport(sp, k, &xerr_idx)) { 2625 XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH); 2626 goto reject; 2627 } 2628 2629 xfrm_pols_put(pols, npols); 2630 return 1; 2631 } 2632 XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK); 2633 2634 reject: 2635 xfrm_secpath_reject(xerr_idx, skb, &fl); 2636 reject_error: 2637 xfrm_pols_put(pols, npols); 2638 return 0; 2639 } 2640 EXPORT_SYMBOL(__xfrm_policy_check); 2641 2642 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family) 2643 { 2644 struct net *net = dev_net(skb->dev); 2645 struct flowi fl; 2646 struct dst_entry *dst; 2647 int res = 1; 2648 2649 if (xfrm_decode_session(skb, &fl, family) < 0) { 2650 XFRM_INC_STATS(net, LINUX_MIB_XFRMFWDHDRERROR); 2651 return 0; 2652 } 2653 2654 skb_dst_force(skb); 2655 2656 dst = xfrm_lookup(net, skb_dst(skb), &fl, NULL, XFRM_LOOKUP_QUEUE); 2657 if (IS_ERR(dst)) { 2658 res = 0; 2659 dst = NULL; 2660 } 2661 skb_dst_set(skb, dst); 2662 return res; 2663 } 2664 EXPORT_SYMBOL(__xfrm_route_forward); 2665 2666 /* Optimize later using cookies and generation ids. */ 2667 2668 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie) 2669 { 2670 /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete 2671 * to DST_OBSOLETE_FORCE_CHK to force all XFRM destinations to 2672 * get validated by dst_ops->check on every use. We do this 2673 * because when a normal route referenced by an XFRM dst is 2674 * obsoleted we do not go looking around for all parent 2675 * referencing XFRM dsts so that we can invalidate them. It 2676 * is just too much work. Instead we make the checks here on 2677 * every use. For example: 2678 * 2679 * XFRM dst A --> IPv4 dst X 2680 * 2681 * X is the "xdst->route" of A (X is also the "dst->path" of A 2682 * in this example). If X is marked obsolete, "A" will not 2683 * notice. That's what we are validating here via the 2684 * stale_bundle() check. 2685 * 2686 * When a policy's bundle is pruned, we dst_free() the XFRM 2687 * dst which causes it's ->obsolete field to be set to 2688 * DST_OBSOLETE_DEAD. If an XFRM dst has been pruned like 2689 * this, we want to force a new route lookup. 2690 */ 2691 if (dst->obsolete < 0 && !stale_bundle(dst)) 2692 return dst; 2693 2694 return NULL; 2695 } 2696 2697 static int stale_bundle(struct dst_entry *dst) 2698 { 2699 return !xfrm_bundle_ok((struct xfrm_dst *)dst); 2700 } 2701 2702 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev) 2703 { 2704 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) { 2705 dst->dev = dev_net(dev)->loopback_dev; 2706 dev_hold(dst->dev); 2707 dev_put(dev); 2708 } 2709 } 2710 EXPORT_SYMBOL(xfrm_dst_ifdown); 2711 2712 static void xfrm_link_failure(struct sk_buff *skb) 2713 { 2714 /* Impossible. Such dst must be popped before reaches point of failure. */ 2715 } 2716 2717 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst) 2718 { 2719 if (dst) { 2720 if (dst->obsolete) { 2721 dst_release(dst); 2722 dst = NULL; 2723 } 2724 } 2725 return dst; 2726 } 2727 2728 void xfrm_garbage_collect(struct net *net) 2729 { 2730 flow_cache_flush(net); 2731 } 2732 EXPORT_SYMBOL(xfrm_garbage_collect); 2733 2734 void xfrm_garbage_collect_deferred(struct net *net) 2735 { 2736 flow_cache_flush_deferred(net); 2737 } 2738 EXPORT_SYMBOL(xfrm_garbage_collect_deferred); 2739 2740 static void xfrm_init_pmtu(struct dst_entry *dst) 2741 { 2742 do { 2743 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 2744 u32 pmtu, route_mtu_cached; 2745 2746 pmtu = dst_mtu(dst->child); 2747 xdst->child_mtu_cached = pmtu; 2748 2749 pmtu = xfrm_state_mtu(dst->xfrm, pmtu); 2750 2751 route_mtu_cached = dst_mtu(xdst->route); 2752 xdst->route_mtu_cached = route_mtu_cached; 2753 2754 if (pmtu > route_mtu_cached) 2755 pmtu = route_mtu_cached; 2756 2757 dst_metric_set(dst, RTAX_MTU, pmtu); 2758 } while ((dst = dst->next)); 2759 } 2760 2761 /* Check that the bundle accepts the flow and its components are 2762 * still valid. 2763 */ 2764 2765 static int xfrm_bundle_ok(struct xfrm_dst *first) 2766 { 2767 struct dst_entry *dst = &first->u.dst; 2768 struct xfrm_dst *last; 2769 u32 mtu; 2770 2771 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) || 2772 (dst->dev && !netif_running(dst->dev))) 2773 return 0; 2774 2775 if (dst->flags & DST_XFRM_QUEUE) 2776 return 1; 2777 2778 last = NULL; 2779 2780 do { 2781 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 2782 2783 if (dst->xfrm->km.state != XFRM_STATE_VALID) 2784 return 0; 2785 if (xdst->xfrm_genid != dst->xfrm->genid) 2786 return 0; 2787 if (xdst->num_pols > 0 && 2788 xdst->policy_genid != atomic_read(&xdst->pols[0]->genid)) 2789 return 0; 2790 2791 mtu = dst_mtu(dst->child); 2792 if (xdst->child_mtu_cached != mtu) { 2793 last = xdst; 2794 xdst->child_mtu_cached = mtu; 2795 } 2796 2797 if (!dst_check(xdst->route, xdst->route_cookie)) 2798 return 0; 2799 mtu = dst_mtu(xdst->route); 2800 if (xdst->route_mtu_cached != mtu) { 2801 last = xdst; 2802 xdst->route_mtu_cached = mtu; 2803 } 2804 2805 dst = dst->child; 2806 } while (dst->xfrm); 2807 2808 if (likely(!last)) 2809 return 1; 2810 2811 mtu = last->child_mtu_cached; 2812 for (;;) { 2813 dst = &last->u.dst; 2814 2815 mtu = xfrm_state_mtu(dst->xfrm, mtu); 2816 if (mtu > last->route_mtu_cached) 2817 mtu = last->route_mtu_cached; 2818 dst_metric_set(dst, RTAX_MTU, mtu); 2819 2820 if (last == first) 2821 break; 2822 2823 last = (struct xfrm_dst *)last->u.dst.next; 2824 last->child_mtu_cached = mtu; 2825 } 2826 2827 return 1; 2828 } 2829 2830 static unsigned int xfrm_default_advmss(const struct dst_entry *dst) 2831 { 2832 return dst_metric_advmss(dst->path); 2833 } 2834 2835 static unsigned int xfrm_mtu(const struct dst_entry *dst) 2836 { 2837 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 2838 2839 return mtu ? : dst_mtu(dst->path); 2840 } 2841 2842 static const void *xfrm_get_dst_nexthop(const struct dst_entry *dst, 2843 const void *daddr) 2844 { 2845 const struct dst_entry *path = dst->path; 2846 2847 for (; dst != path; dst = dst->child) { 2848 const struct xfrm_state *xfrm = dst->xfrm; 2849 2850 if (xfrm->props.mode == XFRM_MODE_TRANSPORT) 2851 continue; 2852 if (xfrm->type->flags & XFRM_TYPE_REMOTE_COADDR) 2853 daddr = xfrm->coaddr; 2854 else if (!(xfrm->type->flags & XFRM_TYPE_LOCAL_COADDR)) 2855 daddr = &xfrm->id.daddr; 2856 } 2857 return daddr; 2858 } 2859 2860 static struct neighbour *xfrm_neigh_lookup(const struct dst_entry *dst, 2861 struct sk_buff *skb, 2862 const void *daddr) 2863 { 2864 const struct dst_entry *path = dst->path; 2865 2866 if (!skb) 2867 daddr = xfrm_get_dst_nexthop(dst, daddr); 2868 return path->ops->neigh_lookup(path, skb, daddr); 2869 } 2870 2871 static void xfrm_confirm_neigh(const struct dst_entry *dst, const void *daddr) 2872 { 2873 const struct dst_entry *path = dst->path; 2874 2875 daddr = xfrm_get_dst_nexthop(dst, daddr); 2876 path->ops->confirm_neigh(path, daddr); 2877 } 2878 2879 int xfrm_policy_register_afinfo(const struct xfrm_policy_afinfo *afinfo, int family) 2880 { 2881 int err = 0; 2882 2883 if (WARN_ON(family >= ARRAY_SIZE(xfrm_policy_afinfo))) 2884 return -EAFNOSUPPORT; 2885 2886 spin_lock(&xfrm_policy_afinfo_lock); 2887 if (unlikely(xfrm_policy_afinfo[family] != NULL)) 2888 err = -EEXIST; 2889 else { 2890 struct dst_ops *dst_ops = afinfo->dst_ops; 2891 if (likely(dst_ops->kmem_cachep == NULL)) 2892 dst_ops->kmem_cachep = xfrm_dst_cache; 2893 if (likely(dst_ops->check == NULL)) 2894 dst_ops->check = xfrm_dst_check; 2895 if (likely(dst_ops->default_advmss == NULL)) 2896 dst_ops->default_advmss = xfrm_default_advmss; 2897 if (likely(dst_ops->mtu == NULL)) 2898 dst_ops->mtu = xfrm_mtu; 2899 if (likely(dst_ops->negative_advice == NULL)) 2900 dst_ops->negative_advice = xfrm_negative_advice; 2901 if (likely(dst_ops->link_failure == NULL)) 2902 dst_ops->link_failure = xfrm_link_failure; 2903 if (likely(dst_ops->neigh_lookup == NULL)) 2904 dst_ops->neigh_lookup = xfrm_neigh_lookup; 2905 if (likely(!dst_ops->confirm_neigh)) 2906 dst_ops->confirm_neigh = xfrm_confirm_neigh; 2907 rcu_assign_pointer(xfrm_policy_afinfo[family], afinfo); 2908 } 2909 spin_unlock(&xfrm_policy_afinfo_lock); 2910 2911 return err; 2912 } 2913 EXPORT_SYMBOL(xfrm_policy_register_afinfo); 2914 2915 void xfrm_policy_unregister_afinfo(const struct xfrm_policy_afinfo *afinfo) 2916 { 2917 struct dst_ops *dst_ops = afinfo->dst_ops; 2918 int i; 2919 2920 for (i = 0; i < ARRAY_SIZE(xfrm_policy_afinfo); i++) { 2921 if (xfrm_policy_afinfo[i] != afinfo) 2922 continue; 2923 RCU_INIT_POINTER(xfrm_policy_afinfo[i], NULL); 2924 break; 2925 } 2926 2927 synchronize_rcu(); 2928 2929 dst_ops->kmem_cachep = NULL; 2930 dst_ops->check = NULL; 2931 dst_ops->negative_advice = NULL; 2932 dst_ops->link_failure = NULL; 2933 } 2934 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo); 2935 2936 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr) 2937 { 2938 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 2939 2940 switch (event) { 2941 case NETDEV_DOWN: 2942 xfrm_garbage_collect(dev_net(dev)); 2943 } 2944 return NOTIFY_DONE; 2945 } 2946 2947 static struct notifier_block xfrm_dev_notifier = { 2948 .notifier_call = xfrm_dev_event, 2949 }; 2950 2951 #ifdef CONFIG_XFRM_STATISTICS 2952 static int __net_init xfrm_statistics_init(struct net *net) 2953 { 2954 int rv; 2955 net->mib.xfrm_statistics = alloc_percpu(struct linux_xfrm_mib); 2956 if (!net->mib.xfrm_statistics) 2957 return -ENOMEM; 2958 rv = xfrm_proc_init(net); 2959 if (rv < 0) 2960 free_percpu(net->mib.xfrm_statistics); 2961 return rv; 2962 } 2963 2964 static void xfrm_statistics_fini(struct net *net) 2965 { 2966 xfrm_proc_fini(net); 2967 free_percpu(net->mib.xfrm_statistics); 2968 } 2969 #else 2970 static int __net_init xfrm_statistics_init(struct net *net) 2971 { 2972 return 0; 2973 } 2974 2975 static void xfrm_statistics_fini(struct net *net) 2976 { 2977 } 2978 #endif 2979 2980 static int __net_init xfrm_policy_init(struct net *net) 2981 { 2982 unsigned int hmask, sz; 2983 int dir; 2984 2985 if (net_eq(net, &init_net)) 2986 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache", 2987 sizeof(struct xfrm_dst), 2988 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, 2989 NULL); 2990 2991 hmask = 8 - 1; 2992 sz = (hmask+1) * sizeof(struct hlist_head); 2993 2994 net->xfrm.policy_byidx = xfrm_hash_alloc(sz); 2995 if (!net->xfrm.policy_byidx) 2996 goto out_byidx; 2997 net->xfrm.policy_idx_hmask = hmask; 2998 2999 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 3000 struct xfrm_policy_hash *htab; 3001 3002 net->xfrm.policy_count[dir] = 0; 3003 net->xfrm.policy_count[XFRM_POLICY_MAX + dir] = 0; 3004 INIT_HLIST_HEAD(&net->xfrm.policy_inexact[dir]); 3005 3006 htab = &net->xfrm.policy_bydst[dir]; 3007 htab->table = xfrm_hash_alloc(sz); 3008 if (!htab->table) 3009 goto out_bydst; 3010 htab->hmask = hmask; 3011 htab->dbits4 = 32; 3012 htab->sbits4 = 32; 3013 htab->dbits6 = 128; 3014 htab->sbits6 = 128; 3015 } 3016 net->xfrm.policy_hthresh.lbits4 = 32; 3017 net->xfrm.policy_hthresh.rbits4 = 32; 3018 net->xfrm.policy_hthresh.lbits6 = 128; 3019 net->xfrm.policy_hthresh.rbits6 = 128; 3020 3021 seqlock_init(&net->xfrm.policy_hthresh.lock); 3022 3023 INIT_LIST_HEAD(&net->xfrm.policy_all); 3024 INIT_WORK(&net->xfrm.policy_hash_work, xfrm_hash_resize); 3025 INIT_WORK(&net->xfrm.policy_hthresh.work, xfrm_hash_rebuild); 3026 if (net_eq(net, &init_net)) 3027 register_netdevice_notifier(&xfrm_dev_notifier); 3028 return 0; 3029 3030 out_bydst: 3031 for (dir--; dir >= 0; dir--) { 3032 struct xfrm_policy_hash *htab; 3033 3034 htab = &net->xfrm.policy_bydst[dir]; 3035 xfrm_hash_free(htab->table, sz); 3036 } 3037 xfrm_hash_free(net->xfrm.policy_byidx, sz); 3038 out_byidx: 3039 return -ENOMEM; 3040 } 3041 3042 static void xfrm_policy_fini(struct net *net) 3043 { 3044 unsigned int sz; 3045 int dir; 3046 3047 flush_work(&net->xfrm.policy_hash_work); 3048 #ifdef CONFIG_XFRM_SUB_POLICY 3049 xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, false); 3050 #endif 3051 xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, false); 3052 3053 WARN_ON(!list_empty(&net->xfrm.policy_all)); 3054 3055 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 3056 struct xfrm_policy_hash *htab; 3057 3058 WARN_ON(!hlist_empty(&net->xfrm.policy_inexact[dir])); 3059 3060 htab = &net->xfrm.policy_bydst[dir]; 3061 sz = (htab->hmask + 1) * sizeof(struct hlist_head); 3062 WARN_ON(!hlist_empty(htab->table)); 3063 xfrm_hash_free(htab->table, sz); 3064 } 3065 3066 sz = (net->xfrm.policy_idx_hmask + 1) * sizeof(struct hlist_head); 3067 WARN_ON(!hlist_empty(net->xfrm.policy_byidx)); 3068 xfrm_hash_free(net->xfrm.policy_byidx, sz); 3069 } 3070 3071 static int __net_init xfrm_net_init(struct net *net) 3072 { 3073 int rv; 3074 3075 /* Initialize the per-net locks here */ 3076 spin_lock_init(&net->xfrm.xfrm_state_lock); 3077 spin_lock_init(&net->xfrm.xfrm_policy_lock); 3078 mutex_init(&net->xfrm.xfrm_cfg_mutex); 3079 3080 rv = xfrm_statistics_init(net); 3081 if (rv < 0) 3082 goto out_statistics; 3083 rv = xfrm_state_init(net); 3084 if (rv < 0) 3085 goto out_state; 3086 rv = xfrm_policy_init(net); 3087 if (rv < 0) 3088 goto out_policy; 3089 rv = xfrm_sysctl_init(net); 3090 if (rv < 0) 3091 goto out_sysctl; 3092 rv = flow_cache_init(net); 3093 if (rv < 0) 3094 goto out; 3095 3096 return 0; 3097 3098 out: 3099 xfrm_sysctl_fini(net); 3100 out_sysctl: 3101 xfrm_policy_fini(net); 3102 out_policy: 3103 xfrm_state_fini(net); 3104 out_state: 3105 xfrm_statistics_fini(net); 3106 out_statistics: 3107 return rv; 3108 } 3109 3110 static void __net_exit xfrm_net_exit(struct net *net) 3111 { 3112 flow_cache_fini(net); 3113 xfrm_sysctl_fini(net); 3114 xfrm_policy_fini(net); 3115 xfrm_state_fini(net); 3116 xfrm_statistics_fini(net); 3117 } 3118 3119 static struct pernet_operations __net_initdata xfrm_net_ops = { 3120 .init = xfrm_net_init, 3121 .exit = xfrm_net_exit, 3122 }; 3123 3124 void __init xfrm_init(void) 3125 { 3126 flow_cache_hp_init(); 3127 register_pernet_subsys(&xfrm_net_ops); 3128 seqcount_init(&xfrm_policy_hash_generation); 3129 xfrm_input_init(); 3130 } 3131 3132 #ifdef CONFIG_AUDITSYSCALL 3133 static void xfrm_audit_common_policyinfo(struct xfrm_policy *xp, 3134 struct audit_buffer *audit_buf) 3135 { 3136 struct xfrm_sec_ctx *ctx = xp->security; 3137 struct xfrm_selector *sel = &xp->selector; 3138 3139 if (ctx) 3140 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s", 3141 ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str); 3142 3143 switch (sel->family) { 3144 case AF_INET: 3145 audit_log_format(audit_buf, " src=%pI4", &sel->saddr.a4); 3146 if (sel->prefixlen_s != 32) 3147 audit_log_format(audit_buf, " src_prefixlen=%d", 3148 sel->prefixlen_s); 3149 audit_log_format(audit_buf, " dst=%pI4", &sel->daddr.a4); 3150 if (sel->prefixlen_d != 32) 3151 audit_log_format(audit_buf, " dst_prefixlen=%d", 3152 sel->prefixlen_d); 3153 break; 3154 case AF_INET6: 3155 audit_log_format(audit_buf, " src=%pI6", sel->saddr.a6); 3156 if (sel->prefixlen_s != 128) 3157 audit_log_format(audit_buf, " src_prefixlen=%d", 3158 sel->prefixlen_s); 3159 audit_log_format(audit_buf, " dst=%pI6", sel->daddr.a6); 3160 if (sel->prefixlen_d != 128) 3161 audit_log_format(audit_buf, " dst_prefixlen=%d", 3162 sel->prefixlen_d); 3163 break; 3164 } 3165 } 3166 3167 void xfrm_audit_policy_add(struct xfrm_policy *xp, int result, bool task_valid) 3168 { 3169 struct audit_buffer *audit_buf; 3170 3171 audit_buf = xfrm_audit_start("SPD-add"); 3172 if (audit_buf == NULL) 3173 return; 3174 xfrm_audit_helper_usrinfo(task_valid, audit_buf); 3175 audit_log_format(audit_buf, " res=%u", result); 3176 xfrm_audit_common_policyinfo(xp, audit_buf); 3177 audit_log_end(audit_buf); 3178 } 3179 EXPORT_SYMBOL_GPL(xfrm_audit_policy_add); 3180 3181 void xfrm_audit_policy_delete(struct xfrm_policy *xp, int result, 3182 bool task_valid) 3183 { 3184 struct audit_buffer *audit_buf; 3185 3186 audit_buf = xfrm_audit_start("SPD-delete"); 3187 if (audit_buf == NULL) 3188 return; 3189 xfrm_audit_helper_usrinfo(task_valid, audit_buf); 3190 audit_log_format(audit_buf, " res=%u", result); 3191 xfrm_audit_common_policyinfo(xp, audit_buf); 3192 audit_log_end(audit_buf); 3193 } 3194 EXPORT_SYMBOL_GPL(xfrm_audit_policy_delete); 3195 #endif 3196 3197 #ifdef CONFIG_XFRM_MIGRATE 3198 static bool xfrm_migrate_selector_match(const struct xfrm_selector *sel_cmp, 3199 const struct xfrm_selector *sel_tgt) 3200 { 3201 if (sel_cmp->proto == IPSEC_ULPROTO_ANY) { 3202 if (sel_tgt->family == sel_cmp->family && 3203 xfrm_addr_equal(&sel_tgt->daddr, &sel_cmp->daddr, 3204 sel_cmp->family) && 3205 xfrm_addr_equal(&sel_tgt->saddr, &sel_cmp->saddr, 3206 sel_cmp->family) && 3207 sel_tgt->prefixlen_d == sel_cmp->prefixlen_d && 3208 sel_tgt->prefixlen_s == sel_cmp->prefixlen_s) { 3209 return true; 3210 } 3211 } else { 3212 if (memcmp(sel_tgt, sel_cmp, sizeof(*sel_tgt)) == 0) { 3213 return true; 3214 } 3215 } 3216 return false; 3217 } 3218 3219 static struct xfrm_policy *xfrm_migrate_policy_find(const struct xfrm_selector *sel, 3220 u8 dir, u8 type, struct net *net) 3221 { 3222 struct xfrm_policy *pol, *ret = NULL; 3223 struct hlist_head *chain; 3224 u32 priority = ~0U; 3225 3226 spin_lock_bh(&net->xfrm.xfrm_policy_lock); 3227 chain = policy_hash_direct(net, &sel->daddr, &sel->saddr, sel->family, dir); 3228 hlist_for_each_entry(pol, chain, bydst) { 3229 if (xfrm_migrate_selector_match(sel, &pol->selector) && 3230 pol->type == type) { 3231 ret = pol; 3232 priority = ret->priority; 3233 break; 3234 } 3235 } 3236 chain = &net->xfrm.policy_inexact[dir]; 3237 hlist_for_each_entry(pol, chain, bydst) { 3238 if ((pol->priority >= priority) && ret) 3239 break; 3240 3241 if (xfrm_migrate_selector_match(sel, &pol->selector) && 3242 pol->type == type) { 3243 ret = pol; 3244 break; 3245 } 3246 } 3247 3248 xfrm_pol_hold(ret); 3249 3250 spin_unlock_bh(&net->xfrm.xfrm_policy_lock); 3251 3252 return ret; 3253 } 3254 3255 static int migrate_tmpl_match(const struct xfrm_migrate *m, const struct xfrm_tmpl *t) 3256 { 3257 int match = 0; 3258 3259 if (t->mode == m->mode && t->id.proto == m->proto && 3260 (m->reqid == 0 || t->reqid == m->reqid)) { 3261 switch (t->mode) { 3262 case XFRM_MODE_TUNNEL: 3263 case XFRM_MODE_BEET: 3264 if (xfrm_addr_equal(&t->id.daddr, &m->old_daddr, 3265 m->old_family) && 3266 xfrm_addr_equal(&t->saddr, &m->old_saddr, 3267 m->old_family)) { 3268 match = 1; 3269 } 3270 break; 3271 case XFRM_MODE_TRANSPORT: 3272 /* in case of transport mode, template does not store 3273 any IP addresses, hence we just compare mode and 3274 protocol */ 3275 match = 1; 3276 break; 3277 default: 3278 break; 3279 } 3280 } 3281 return match; 3282 } 3283 3284 /* update endpoint address(es) of template(s) */ 3285 static int xfrm_policy_migrate(struct xfrm_policy *pol, 3286 struct xfrm_migrate *m, int num_migrate) 3287 { 3288 struct xfrm_migrate *mp; 3289 int i, j, n = 0; 3290 3291 write_lock_bh(&pol->lock); 3292 if (unlikely(pol->walk.dead)) { 3293 /* target policy has been deleted */ 3294 write_unlock_bh(&pol->lock); 3295 return -ENOENT; 3296 } 3297 3298 for (i = 0; i < pol->xfrm_nr; i++) { 3299 for (j = 0, mp = m; j < num_migrate; j++, mp++) { 3300 if (!migrate_tmpl_match(mp, &pol->xfrm_vec[i])) 3301 continue; 3302 n++; 3303 if (pol->xfrm_vec[i].mode != XFRM_MODE_TUNNEL && 3304 pol->xfrm_vec[i].mode != XFRM_MODE_BEET) 3305 continue; 3306 /* update endpoints */ 3307 memcpy(&pol->xfrm_vec[i].id.daddr, &mp->new_daddr, 3308 sizeof(pol->xfrm_vec[i].id.daddr)); 3309 memcpy(&pol->xfrm_vec[i].saddr, &mp->new_saddr, 3310 sizeof(pol->xfrm_vec[i].saddr)); 3311 pol->xfrm_vec[i].encap_family = mp->new_family; 3312 /* flush bundles */ 3313 atomic_inc(&pol->genid); 3314 } 3315 } 3316 3317 write_unlock_bh(&pol->lock); 3318 3319 if (!n) 3320 return -ENODATA; 3321 3322 return 0; 3323 } 3324 3325 static int xfrm_migrate_check(const struct xfrm_migrate *m, int num_migrate) 3326 { 3327 int i, j; 3328 3329 if (num_migrate < 1 || num_migrate > XFRM_MAX_DEPTH) 3330 return -EINVAL; 3331 3332 for (i = 0; i < num_migrate; i++) { 3333 if (xfrm_addr_equal(&m[i].old_daddr, &m[i].new_daddr, 3334 m[i].old_family) && 3335 xfrm_addr_equal(&m[i].old_saddr, &m[i].new_saddr, 3336 m[i].old_family)) 3337 return -EINVAL; 3338 if (xfrm_addr_any(&m[i].new_daddr, m[i].new_family) || 3339 xfrm_addr_any(&m[i].new_saddr, m[i].new_family)) 3340 return -EINVAL; 3341 3342 /* check if there is any duplicated entry */ 3343 for (j = i + 1; j < num_migrate; j++) { 3344 if (!memcmp(&m[i].old_daddr, &m[j].old_daddr, 3345 sizeof(m[i].old_daddr)) && 3346 !memcmp(&m[i].old_saddr, &m[j].old_saddr, 3347 sizeof(m[i].old_saddr)) && 3348 m[i].proto == m[j].proto && 3349 m[i].mode == m[j].mode && 3350 m[i].reqid == m[j].reqid && 3351 m[i].old_family == m[j].old_family) 3352 return -EINVAL; 3353 } 3354 } 3355 3356 return 0; 3357 } 3358 3359 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type, 3360 struct xfrm_migrate *m, int num_migrate, 3361 struct xfrm_kmaddress *k, struct net *net) 3362 { 3363 int i, err, nx_cur = 0, nx_new = 0; 3364 struct xfrm_policy *pol = NULL; 3365 struct xfrm_state *x, *xc; 3366 struct xfrm_state *x_cur[XFRM_MAX_DEPTH]; 3367 struct xfrm_state *x_new[XFRM_MAX_DEPTH]; 3368 struct xfrm_migrate *mp; 3369 3370 if ((err = xfrm_migrate_check(m, num_migrate)) < 0) 3371 goto out; 3372 3373 /* Stage 1 - find policy */ 3374 if ((pol = xfrm_migrate_policy_find(sel, dir, type, net)) == NULL) { 3375 err = -ENOENT; 3376 goto out; 3377 } 3378 3379 /* Stage 2 - find and update state(s) */ 3380 for (i = 0, mp = m; i < num_migrate; i++, mp++) { 3381 if ((x = xfrm_migrate_state_find(mp, net))) { 3382 x_cur[nx_cur] = x; 3383 nx_cur++; 3384 if ((xc = xfrm_state_migrate(x, mp))) { 3385 x_new[nx_new] = xc; 3386 nx_new++; 3387 } else { 3388 err = -ENODATA; 3389 goto restore_state; 3390 } 3391 } 3392 } 3393 3394 /* Stage 3 - update policy */ 3395 if ((err = xfrm_policy_migrate(pol, m, num_migrate)) < 0) 3396 goto restore_state; 3397 3398 /* Stage 4 - delete old state(s) */ 3399 if (nx_cur) { 3400 xfrm_states_put(x_cur, nx_cur); 3401 xfrm_states_delete(x_cur, nx_cur); 3402 } 3403 3404 /* Stage 5 - announce */ 3405 km_migrate(sel, dir, type, m, num_migrate, k); 3406 3407 xfrm_pol_put(pol); 3408 3409 return 0; 3410 out: 3411 return err; 3412 3413 restore_state: 3414 if (pol) 3415 xfrm_pol_put(pol); 3416 if (nx_cur) 3417 xfrm_states_put(x_cur, nx_cur); 3418 if (nx_new) 3419 xfrm_states_delete(x_new, nx_new); 3420 3421 return err; 3422 } 3423 EXPORT_SYMBOL(xfrm_migrate); 3424 #endif 3425