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/config.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 <net/xfrm.h> 27 #include <net/ip.h> 28 29 DECLARE_MUTEX(xfrm_cfg_sem); 30 EXPORT_SYMBOL(xfrm_cfg_sem); 31 32 static DEFINE_RWLOCK(xfrm_policy_lock); 33 34 struct xfrm_policy *xfrm_policy_list[XFRM_POLICY_MAX*2]; 35 EXPORT_SYMBOL(xfrm_policy_list); 36 37 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock); 38 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO]; 39 40 static kmem_cache_t *xfrm_dst_cache __read_mostly; 41 42 static struct work_struct xfrm_policy_gc_work; 43 static struct list_head xfrm_policy_gc_list = 44 LIST_HEAD_INIT(xfrm_policy_gc_list); 45 static DEFINE_SPINLOCK(xfrm_policy_gc_lock); 46 47 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family); 48 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo); 49 50 int xfrm_register_type(struct xfrm_type *type, unsigned short family) 51 { 52 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 53 struct xfrm_type_map *typemap; 54 int err = 0; 55 56 if (unlikely(afinfo == NULL)) 57 return -EAFNOSUPPORT; 58 typemap = afinfo->type_map; 59 60 write_lock(&typemap->lock); 61 if (likely(typemap->map[type->proto] == NULL)) 62 typemap->map[type->proto] = type; 63 else 64 err = -EEXIST; 65 write_unlock(&typemap->lock); 66 xfrm_policy_put_afinfo(afinfo); 67 return err; 68 } 69 EXPORT_SYMBOL(xfrm_register_type); 70 71 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family) 72 { 73 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 74 struct xfrm_type_map *typemap; 75 int err = 0; 76 77 if (unlikely(afinfo == NULL)) 78 return -EAFNOSUPPORT; 79 typemap = afinfo->type_map; 80 81 write_lock(&typemap->lock); 82 if (unlikely(typemap->map[type->proto] != type)) 83 err = -ENOENT; 84 else 85 typemap->map[type->proto] = NULL; 86 write_unlock(&typemap->lock); 87 xfrm_policy_put_afinfo(afinfo); 88 return err; 89 } 90 EXPORT_SYMBOL(xfrm_unregister_type); 91 92 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family) 93 { 94 struct xfrm_policy_afinfo *afinfo; 95 struct xfrm_type_map *typemap; 96 struct xfrm_type *type; 97 int modload_attempted = 0; 98 99 retry: 100 afinfo = xfrm_policy_get_afinfo(family); 101 if (unlikely(afinfo == NULL)) 102 return NULL; 103 typemap = afinfo->type_map; 104 105 read_lock(&typemap->lock); 106 type = typemap->map[proto]; 107 if (unlikely(type && !try_module_get(type->owner))) 108 type = NULL; 109 read_unlock(&typemap->lock); 110 if (!type && !modload_attempted) { 111 xfrm_policy_put_afinfo(afinfo); 112 request_module("xfrm-type-%d-%d", 113 (int) family, (int) proto); 114 modload_attempted = 1; 115 goto retry; 116 } 117 118 xfrm_policy_put_afinfo(afinfo); 119 return type; 120 } 121 122 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl, 123 unsigned short family) 124 { 125 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 126 int err = 0; 127 128 if (unlikely(afinfo == NULL)) 129 return -EAFNOSUPPORT; 130 131 if (likely(afinfo->dst_lookup != NULL)) 132 err = afinfo->dst_lookup(dst, fl); 133 else 134 err = -EINVAL; 135 xfrm_policy_put_afinfo(afinfo); 136 return err; 137 } 138 EXPORT_SYMBOL(xfrm_dst_lookup); 139 140 void xfrm_put_type(struct xfrm_type *type) 141 { 142 module_put(type->owner); 143 } 144 145 static inline unsigned long make_jiffies(long secs) 146 { 147 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ) 148 return MAX_SCHEDULE_TIMEOUT-1; 149 else 150 return secs*HZ; 151 } 152 153 static void xfrm_policy_timer(unsigned long data) 154 { 155 struct xfrm_policy *xp = (struct xfrm_policy*)data; 156 unsigned long now = (unsigned long)xtime.tv_sec; 157 long next = LONG_MAX; 158 int warn = 0; 159 int dir; 160 161 read_lock(&xp->lock); 162 163 if (xp->dead) 164 goto out; 165 166 dir = xfrm_policy_id2dir(xp->index); 167 168 if (xp->lft.hard_add_expires_seconds) { 169 long tmo = xp->lft.hard_add_expires_seconds + 170 xp->curlft.add_time - now; 171 if (tmo <= 0) 172 goto expired; 173 if (tmo < next) 174 next = tmo; 175 } 176 if (xp->lft.hard_use_expires_seconds) { 177 long tmo = xp->lft.hard_use_expires_seconds + 178 (xp->curlft.use_time ? : xp->curlft.add_time) - now; 179 if (tmo <= 0) 180 goto expired; 181 if (tmo < next) 182 next = tmo; 183 } 184 if (xp->lft.soft_add_expires_seconds) { 185 long tmo = xp->lft.soft_add_expires_seconds + 186 xp->curlft.add_time - now; 187 if (tmo <= 0) { 188 warn = 1; 189 tmo = XFRM_KM_TIMEOUT; 190 } 191 if (tmo < next) 192 next = tmo; 193 } 194 if (xp->lft.soft_use_expires_seconds) { 195 long tmo = xp->lft.soft_use_expires_seconds + 196 (xp->curlft.use_time ? : xp->curlft.add_time) - now; 197 if (tmo <= 0) { 198 warn = 1; 199 tmo = XFRM_KM_TIMEOUT; 200 } 201 if (tmo < next) 202 next = tmo; 203 } 204 205 if (warn) 206 km_policy_expired(xp, dir, 0); 207 if (next != LONG_MAX && 208 !mod_timer(&xp->timer, jiffies + make_jiffies(next))) 209 xfrm_pol_hold(xp); 210 211 out: 212 read_unlock(&xp->lock); 213 xfrm_pol_put(xp); 214 return; 215 216 expired: 217 read_unlock(&xp->lock); 218 if (!xfrm_policy_delete(xp, dir)) 219 km_policy_expired(xp, dir, 1); 220 xfrm_pol_put(xp); 221 } 222 223 224 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2 225 * SPD calls. 226 */ 227 228 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp) 229 { 230 struct xfrm_policy *policy; 231 232 policy = kmalloc(sizeof(struct xfrm_policy), gfp); 233 234 if (policy) { 235 memset(policy, 0, sizeof(struct xfrm_policy)); 236 atomic_set(&policy->refcnt, 1); 237 rwlock_init(&policy->lock); 238 init_timer(&policy->timer); 239 policy->timer.data = (unsigned long)policy; 240 policy->timer.function = xfrm_policy_timer; 241 } 242 return policy; 243 } 244 EXPORT_SYMBOL(xfrm_policy_alloc); 245 246 /* Destroy xfrm_policy: descendant resources must be released to this moment. */ 247 248 void __xfrm_policy_destroy(struct xfrm_policy *policy) 249 { 250 BUG_ON(!policy->dead); 251 252 BUG_ON(policy->bundles); 253 254 if (del_timer(&policy->timer)) 255 BUG(); 256 257 security_xfrm_policy_free(policy); 258 kfree(policy); 259 } 260 EXPORT_SYMBOL(__xfrm_policy_destroy); 261 262 static void xfrm_policy_gc_kill(struct xfrm_policy *policy) 263 { 264 struct dst_entry *dst; 265 266 while ((dst = policy->bundles) != NULL) { 267 policy->bundles = dst->next; 268 dst_free(dst); 269 } 270 271 if (del_timer(&policy->timer)) 272 atomic_dec(&policy->refcnt); 273 274 if (atomic_read(&policy->refcnt) > 1) 275 flow_cache_flush(); 276 277 xfrm_pol_put(policy); 278 } 279 280 static void xfrm_policy_gc_task(void *data) 281 { 282 struct xfrm_policy *policy; 283 struct list_head *entry, *tmp; 284 struct list_head gc_list = LIST_HEAD_INIT(gc_list); 285 286 spin_lock_bh(&xfrm_policy_gc_lock); 287 list_splice_init(&xfrm_policy_gc_list, &gc_list); 288 spin_unlock_bh(&xfrm_policy_gc_lock); 289 290 list_for_each_safe(entry, tmp, &gc_list) { 291 policy = list_entry(entry, struct xfrm_policy, list); 292 xfrm_policy_gc_kill(policy); 293 } 294 } 295 296 /* Rule must be locked. Release descentant resources, announce 297 * entry dead. The rule must be unlinked from lists to the moment. 298 */ 299 300 static void xfrm_policy_kill(struct xfrm_policy *policy) 301 { 302 int dead; 303 304 write_lock_bh(&policy->lock); 305 dead = policy->dead; 306 policy->dead = 1; 307 write_unlock_bh(&policy->lock); 308 309 if (unlikely(dead)) { 310 WARN_ON(1); 311 return; 312 } 313 314 spin_lock(&xfrm_policy_gc_lock); 315 list_add(&policy->list, &xfrm_policy_gc_list); 316 spin_unlock(&xfrm_policy_gc_lock); 317 318 schedule_work(&xfrm_policy_gc_work); 319 } 320 321 /* Generate new index... KAME seems to generate them ordered by cost 322 * of an absolute inpredictability of ordering of rules. This will not pass. */ 323 static u32 xfrm_gen_index(int dir) 324 { 325 u32 idx; 326 struct xfrm_policy *p; 327 static u32 idx_generator; 328 329 for (;;) { 330 idx = (idx_generator | dir); 331 idx_generator += 8; 332 if (idx == 0) 333 idx = 8; 334 for (p = xfrm_policy_list[dir]; p; p = p->next) { 335 if (p->index == idx) 336 break; 337 } 338 if (!p) 339 return idx; 340 } 341 } 342 343 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl) 344 { 345 struct xfrm_policy *pol, **p; 346 struct xfrm_policy *delpol = NULL; 347 struct xfrm_policy **newpos = NULL; 348 struct dst_entry *gc_list; 349 350 write_lock_bh(&xfrm_policy_lock); 351 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) { 352 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0 && 353 xfrm_sec_ctx_match(pol->security, policy->security)) { 354 if (excl) { 355 write_unlock_bh(&xfrm_policy_lock); 356 return -EEXIST; 357 } 358 *p = pol->next; 359 delpol = pol; 360 if (policy->priority > pol->priority) 361 continue; 362 } else if (policy->priority >= pol->priority) { 363 p = &pol->next; 364 continue; 365 } 366 if (!newpos) 367 newpos = p; 368 if (delpol) 369 break; 370 p = &pol->next; 371 } 372 if (newpos) 373 p = newpos; 374 xfrm_pol_hold(policy); 375 policy->next = *p; 376 *p = policy; 377 atomic_inc(&flow_cache_genid); 378 policy->index = delpol ? delpol->index : xfrm_gen_index(dir); 379 policy->curlft.add_time = (unsigned long)xtime.tv_sec; 380 policy->curlft.use_time = 0; 381 if (!mod_timer(&policy->timer, jiffies + HZ)) 382 xfrm_pol_hold(policy); 383 write_unlock_bh(&xfrm_policy_lock); 384 385 if (delpol) 386 xfrm_policy_kill(delpol); 387 388 read_lock_bh(&xfrm_policy_lock); 389 gc_list = NULL; 390 for (policy = policy->next; policy; policy = policy->next) { 391 struct dst_entry *dst; 392 393 write_lock(&policy->lock); 394 dst = policy->bundles; 395 if (dst) { 396 struct dst_entry *tail = dst; 397 while (tail->next) 398 tail = tail->next; 399 tail->next = gc_list; 400 gc_list = dst; 401 402 policy->bundles = NULL; 403 } 404 write_unlock(&policy->lock); 405 } 406 read_unlock_bh(&xfrm_policy_lock); 407 408 while (gc_list) { 409 struct dst_entry *dst = gc_list; 410 411 gc_list = dst->next; 412 dst_free(dst); 413 } 414 415 return 0; 416 } 417 EXPORT_SYMBOL(xfrm_policy_insert); 418 419 struct xfrm_policy *xfrm_policy_bysel_ctx(int dir, struct xfrm_selector *sel, 420 struct xfrm_sec_ctx *ctx, int delete) 421 { 422 struct xfrm_policy *pol, **p; 423 424 write_lock_bh(&xfrm_policy_lock); 425 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) { 426 if ((memcmp(sel, &pol->selector, sizeof(*sel)) == 0) && 427 (xfrm_sec_ctx_match(ctx, pol->security))) { 428 xfrm_pol_hold(pol); 429 if (delete) 430 *p = pol->next; 431 break; 432 } 433 } 434 write_unlock_bh(&xfrm_policy_lock); 435 436 if (pol && delete) { 437 atomic_inc(&flow_cache_genid); 438 xfrm_policy_kill(pol); 439 } 440 return pol; 441 } 442 EXPORT_SYMBOL(xfrm_policy_bysel_ctx); 443 444 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete) 445 { 446 struct xfrm_policy *pol, **p; 447 448 write_lock_bh(&xfrm_policy_lock); 449 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) { 450 if (pol->index == id) { 451 xfrm_pol_hold(pol); 452 if (delete) 453 *p = pol->next; 454 break; 455 } 456 } 457 write_unlock_bh(&xfrm_policy_lock); 458 459 if (pol && delete) { 460 atomic_inc(&flow_cache_genid); 461 xfrm_policy_kill(pol); 462 } 463 return pol; 464 } 465 EXPORT_SYMBOL(xfrm_policy_byid); 466 467 void xfrm_policy_flush(void) 468 { 469 struct xfrm_policy *xp; 470 int dir; 471 472 write_lock_bh(&xfrm_policy_lock); 473 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) { 474 while ((xp = xfrm_policy_list[dir]) != NULL) { 475 xfrm_policy_list[dir] = xp->next; 476 write_unlock_bh(&xfrm_policy_lock); 477 478 xfrm_policy_kill(xp); 479 480 write_lock_bh(&xfrm_policy_lock); 481 } 482 } 483 atomic_inc(&flow_cache_genid); 484 write_unlock_bh(&xfrm_policy_lock); 485 } 486 EXPORT_SYMBOL(xfrm_policy_flush); 487 488 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*), 489 void *data) 490 { 491 struct xfrm_policy *xp; 492 int dir; 493 int count = 0; 494 int error = 0; 495 496 read_lock_bh(&xfrm_policy_lock); 497 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) { 498 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) 499 count++; 500 } 501 502 if (count == 0) { 503 error = -ENOENT; 504 goto out; 505 } 506 507 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) { 508 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) { 509 error = func(xp, dir%XFRM_POLICY_MAX, --count, data); 510 if (error) 511 goto out; 512 } 513 } 514 515 out: 516 read_unlock_bh(&xfrm_policy_lock); 517 return error; 518 } 519 EXPORT_SYMBOL(xfrm_policy_walk); 520 521 /* Find policy to apply to this flow. */ 522 523 static void xfrm_policy_lookup(struct flowi *fl, u32 sk_sid, u16 family, u8 dir, 524 void **objp, atomic_t **obj_refp) 525 { 526 struct xfrm_policy *pol; 527 528 read_lock_bh(&xfrm_policy_lock); 529 for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) { 530 struct xfrm_selector *sel = &pol->selector; 531 int match; 532 533 if (pol->family != family) 534 continue; 535 536 match = xfrm_selector_match(sel, fl, family); 537 538 if (match) { 539 if (!security_xfrm_policy_lookup(pol, sk_sid, dir)) { 540 xfrm_pol_hold(pol); 541 break; 542 } 543 } 544 } 545 read_unlock_bh(&xfrm_policy_lock); 546 if ((*objp = (void *) pol) != NULL) 547 *obj_refp = &pol->refcnt; 548 } 549 550 static inline int policy_to_flow_dir(int dir) 551 { 552 if (XFRM_POLICY_IN == FLOW_DIR_IN && 553 XFRM_POLICY_OUT == FLOW_DIR_OUT && 554 XFRM_POLICY_FWD == FLOW_DIR_FWD) 555 return dir; 556 switch (dir) { 557 default: 558 case XFRM_POLICY_IN: 559 return FLOW_DIR_IN; 560 case XFRM_POLICY_OUT: 561 return FLOW_DIR_OUT; 562 case XFRM_POLICY_FWD: 563 return FLOW_DIR_FWD; 564 }; 565 } 566 567 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl, u32 sk_sid) 568 { 569 struct xfrm_policy *pol; 570 571 read_lock_bh(&xfrm_policy_lock); 572 if ((pol = sk->sk_policy[dir]) != NULL) { 573 int match = xfrm_selector_match(&pol->selector, fl, 574 sk->sk_family); 575 int err = 0; 576 577 if (match) 578 err = security_xfrm_policy_lookup(pol, sk_sid, policy_to_flow_dir(dir)); 579 580 if (match && !err) 581 xfrm_pol_hold(pol); 582 else 583 pol = NULL; 584 } 585 read_unlock_bh(&xfrm_policy_lock); 586 return pol; 587 } 588 589 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir) 590 { 591 pol->next = xfrm_policy_list[dir]; 592 xfrm_policy_list[dir] = pol; 593 xfrm_pol_hold(pol); 594 } 595 596 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol, 597 int dir) 598 { 599 struct xfrm_policy **polp; 600 601 for (polp = &xfrm_policy_list[dir]; 602 *polp != NULL; polp = &(*polp)->next) { 603 if (*polp == pol) { 604 *polp = pol->next; 605 return pol; 606 } 607 } 608 return NULL; 609 } 610 611 int xfrm_policy_delete(struct xfrm_policy *pol, int dir) 612 { 613 write_lock_bh(&xfrm_policy_lock); 614 pol = __xfrm_policy_unlink(pol, dir); 615 write_unlock_bh(&xfrm_policy_lock); 616 if (pol) { 617 if (dir < XFRM_POLICY_MAX) 618 atomic_inc(&flow_cache_genid); 619 xfrm_policy_kill(pol); 620 return 0; 621 } 622 return -ENOENT; 623 } 624 625 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol) 626 { 627 struct xfrm_policy *old_pol; 628 629 write_lock_bh(&xfrm_policy_lock); 630 old_pol = sk->sk_policy[dir]; 631 sk->sk_policy[dir] = pol; 632 if (pol) { 633 pol->curlft.add_time = (unsigned long)xtime.tv_sec; 634 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir); 635 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir); 636 } 637 if (old_pol) 638 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir); 639 write_unlock_bh(&xfrm_policy_lock); 640 641 if (old_pol) { 642 xfrm_policy_kill(old_pol); 643 } 644 return 0; 645 } 646 647 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir) 648 { 649 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC); 650 651 if (newp) { 652 newp->selector = old->selector; 653 if (security_xfrm_policy_clone(old, newp)) { 654 kfree(newp); 655 return NULL; /* ENOMEM */ 656 } 657 newp->lft = old->lft; 658 newp->curlft = old->curlft; 659 newp->action = old->action; 660 newp->flags = old->flags; 661 newp->xfrm_nr = old->xfrm_nr; 662 newp->index = old->index; 663 memcpy(newp->xfrm_vec, old->xfrm_vec, 664 newp->xfrm_nr*sizeof(struct xfrm_tmpl)); 665 write_lock_bh(&xfrm_policy_lock); 666 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir); 667 write_unlock_bh(&xfrm_policy_lock); 668 xfrm_pol_put(newp); 669 } 670 return newp; 671 } 672 673 int __xfrm_sk_clone_policy(struct sock *sk) 674 { 675 struct xfrm_policy *p0 = sk->sk_policy[0], 676 *p1 = sk->sk_policy[1]; 677 678 sk->sk_policy[0] = sk->sk_policy[1] = NULL; 679 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL) 680 return -ENOMEM; 681 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL) 682 return -ENOMEM; 683 return 0; 684 } 685 686 /* Resolve list of templates for the flow, given policy. */ 687 688 static int 689 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl, 690 struct xfrm_state **xfrm, 691 unsigned short family) 692 { 693 int nx; 694 int i, error; 695 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family); 696 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family); 697 698 for (nx=0, i = 0; i < policy->xfrm_nr; i++) { 699 struct xfrm_state *x; 700 xfrm_address_t *remote = daddr; 701 xfrm_address_t *local = saddr; 702 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i]; 703 704 if (tmpl->mode) { 705 remote = &tmpl->id.daddr; 706 local = &tmpl->saddr; 707 } 708 709 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family); 710 711 if (x && x->km.state == XFRM_STATE_VALID) { 712 xfrm[nx++] = x; 713 daddr = remote; 714 saddr = local; 715 continue; 716 } 717 if (x) { 718 error = (x->km.state == XFRM_STATE_ERROR ? 719 -EINVAL : -EAGAIN); 720 xfrm_state_put(x); 721 } 722 723 if (!tmpl->optional) 724 goto fail; 725 } 726 return nx; 727 728 fail: 729 for (nx--; nx>=0; nx--) 730 xfrm_state_put(xfrm[nx]); 731 return error; 732 } 733 734 /* Check that the bundle accepts the flow and its components are 735 * still valid. 736 */ 737 738 static struct dst_entry * 739 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family) 740 { 741 struct dst_entry *x; 742 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 743 if (unlikely(afinfo == NULL)) 744 return ERR_PTR(-EINVAL); 745 x = afinfo->find_bundle(fl, policy); 746 xfrm_policy_put_afinfo(afinfo); 747 return x; 748 } 749 750 /* Allocate chain of dst_entry's, attach known xfrm's, calculate 751 * all the metrics... Shortly, bundle a bundle. 752 */ 753 754 static int 755 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx, 756 struct flowi *fl, struct dst_entry **dst_p, 757 unsigned short family) 758 { 759 int err; 760 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 761 if (unlikely(afinfo == NULL)) 762 return -EINVAL; 763 err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p); 764 xfrm_policy_put_afinfo(afinfo); 765 return err; 766 } 767 768 769 static int stale_bundle(struct dst_entry *dst); 770 771 /* Main function: finds/creates a bundle for given flow. 772 * 773 * At the moment we eat a raw IP route. Mostly to speed up lookups 774 * on interfaces with disabled IPsec. 775 */ 776 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl, 777 struct sock *sk, int flags) 778 { 779 struct xfrm_policy *policy; 780 struct xfrm_state *xfrm[XFRM_MAX_DEPTH]; 781 struct dst_entry *dst, *dst_orig = *dst_p; 782 int nx = 0; 783 int err; 784 u32 genid; 785 u16 family; 786 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT); 787 u32 sk_sid = security_sk_sid(sk, fl, dir); 788 restart: 789 genid = atomic_read(&flow_cache_genid); 790 policy = NULL; 791 if (sk && sk->sk_policy[1]) 792 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, sk_sid); 793 794 if (!policy) { 795 /* To accelerate a bit... */ 796 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT]) 797 return 0; 798 799 policy = flow_cache_lookup(fl, sk_sid, dst_orig->ops->family, 800 dir, xfrm_policy_lookup); 801 } 802 803 if (!policy) 804 return 0; 805 806 family = dst_orig->ops->family; 807 policy->curlft.use_time = (unsigned long)xtime.tv_sec; 808 809 switch (policy->action) { 810 case XFRM_POLICY_BLOCK: 811 /* Prohibit the flow */ 812 err = -EPERM; 813 goto error; 814 815 case XFRM_POLICY_ALLOW: 816 if (policy->xfrm_nr == 0) { 817 /* Flow passes not transformed. */ 818 xfrm_pol_put(policy); 819 return 0; 820 } 821 822 /* Try to find matching bundle. 823 * 824 * LATER: help from flow cache. It is optional, this 825 * is required only for output policy. 826 */ 827 dst = xfrm_find_bundle(fl, policy, family); 828 if (IS_ERR(dst)) { 829 err = PTR_ERR(dst); 830 goto error; 831 } 832 833 if (dst) 834 break; 835 836 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family); 837 838 if (unlikely(nx<0)) { 839 err = nx; 840 if (err == -EAGAIN && flags) { 841 DECLARE_WAITQUEUE(wait, current); 842 843 add_wait_queue(&km_waitq, &wait); 844 set_current_state(TASK_INTERRUPTIBLE); 845 schedule(); 846 set_current_state(TASK_RUNNING); 847 remove_wait_queue(&km_waitq, &wait); 848 849 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family); 850 851 if (nx == -EAGAIN && signal_pending(current)) { 852 err = -ERESTART; 853 goto error; 854 } 855 if (nx == -EAGAIN || 856 genid != atomic_read(&flow_cache_genid)) { 857 xfrm_pol_put(policy); 858 goto restart; 859 } 860 err = nx; 861 } 862 if (err < 0) 863 goto error; 864 } 865 if (nx == 0) { 866 /* Flow passes not transformed. */ 867 xfrm_pol_put(policy); 868 return 0; 869 } 870 871 dst = dst_orig; 872 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family); 873 874 if (unlikely(err)) { 875 int i; 876 for (i=0; i<nx; i++) 877 xfrm_state_put(xfrm[i]); 878 goto error; 879 } 880 881 write_lock_bh(&policy->lock); 882 if (unlikely(policy->dead || stale_bundle(dst))) { 883 /* Wow! While we worked on resolving, this 884 * policy has gone. Retry. It is not paranoia, 885 * we just cannot enlist new bundle to dead object. 886 * We can't enlist stable bundles either. 887 */ 888 write_unlock_bh(&policy->lock); 889 if (dst) 890 dst_free(dst); 891 892 err = -EHOSTUNREACH; 893 goto error; 894 } 895 dst->next = policy->bundles; 896 policy->bundles = dst; 897 dst_hold(dst); 898 write_unlock_bh(&policy->lock); 899 } 900 *dst_p = dst; 901 dst_release(dst_orig); 902 xfrm_pol_put(policy); 903 return 0; 904 905 error: 906 dst_release(dst_orig); 907 xfrm_pol_put(policy); 908 *dst_p = NULL; 909 return err; 910 } 911 EXPORT_SYMBOL(xfrm_lookup); 912 913 /* When skb is transformed back to its "native" form, we have to 914 * check policy restrictions. At the moment we make this in maximally 915 * stupid way. Shame on me. :-) Of course, connected sockets must 916 * have policy cached at them. 917 */ 918 919 static inline int 920 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x, 921 unsigned short family) 922 { 923 if (xfrm_state_kern(x)) 924 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family); 925 return x->id.proto == tmpl->id.proto && 926 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) && 927 (x->props.reqid == tmpl->reqid || !tmpl->reqid) && 928 x->props.mode == tmpl->mode && 929 (tmpl->aalgos & (1<<x->props.aalgo)) && 930 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family)); 931 } 932 933 static inline int 934 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start, 935 unsigned short family) 936 { 937 int idx = start; 938 939 if (tmpl->optional) { 940 if (!tmpl->mode) 941 return start; 942 } else 943 start = -1; 944 for (; idx < sp->len; idx++) { 945 if (xfrm_state_ok(tmpl, sp->x[idx].xvec, family)) 946 return ++idx; 947 if (sp->x[idx].xvec->props.mode) 948 break; 949 } 950 return start; 951 } 952 953 int 954 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family) 955 { 956 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family); 957 958 if (unlikely(afinfo == NULL)) 959 return -EAFNOSUPPORT; 960 961 afinfo->decode_session(skb, fl); 962 xfrm_policy_put_afinfo(afinfo); 963 return 0; 964 } 965 EXPORT_SYMBOL(xfrm_decode_session); 966 967 static inline int secpath_has_tunnel(struct sec_path *sp, int k) 968 { 969 for (; k < sp->len; k++) { 970 if (sp->x[k].xvec->props.mode) 971 return 1; 972 } 973 974 return 0; 975 } 976 977 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 978 unsigned short family) 979 { 980 struct xfrm_policy *pol; 981 struct flowi fl; 982 u8 fl_dir = policy_to_flow_dir(dir); 983 u32 sk_sid; 984 985 if (xfrm_decode_session(skb, &fl, family) < 0) 986 return 0; 987 nf_nat_decode_session(skb, &fl, family); 988 989 sk_sid = security_sk_sid(sk, &fl, fl_dir); 990 991 /* First, check used SA against their selectors. */ 992 if (skb->sp) { 993 int i; 994 995 for (i=skb->sp->len-1; i>=0; i--) { 996 struct sec_decap_state *xvec = &(skb->sp->x[i]); 997 if (!xfrm_selector_match(&xvec->xvec->sel, &fl, family)) 998 return 0; 999 } 1000 } 1001 1002 pol = NULL; 1003 if (sk && sk->sk_policy[dir]) 1004 pol = xfrm_sk_policy_lookup(sk, dir, &fl, sk_sid); 1005 1006 if (!pol) 1007 pol = flow_cache_lookup(&fl, sk_sid, family, fl_dir, 1008 xfrm_policy_lookup); 1009 1010 if (!pol) 1011 return !skb->sp || !secpath_has_tunnel(skb->sp, 0); 1012 1013 pol->curlft.use_time = (unsigned long)xtime.tv_sec; 1014 1015 if (pol->action == XFRM_POLICY_ALLOW) { 1016 struct sec_path *sp; 1017 static struct sec_path dummy; 1018 int i, k; 1019 1020 if ((sp = skb->sp) == NULL) 1021 sp = &dummy; 1022 1023 /* For each tunnel xfrm, find the first matching tmpl. 1024 * For each tmpl before that, find corresponding xfrm. 1025 * Order is _important_. Later we will implement 1026 * some barriers, but at the moment barriers 1027 * are implied between each two transformations. 1028 */ 1029 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) { 1030 k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family); 1031 if (k < 0) 1032 goto reject; 1033 } 1034 1035 if (secpath_has_tunnel(sp, k)) 1036 goto reject; 1037 1038 xfrm_pol_put(pol); 1039 return 1; 1040 } 1041 1042 reject: 1043 xfrm_pol_put(pol); 1044 return 0; 1045 } 1046 EXPORT_SYMBOL(__xfrm_policy_check); 1047 1048 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family) 1049 { 1050 struct flowi fl; 1051 1052 if (xfrm_decode_session(skb, &fl, family) < 0) 1053 return 0; 1054 1055 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0; 1056 } 1057 EXPORT_SYMBOL(__xfrm_route_forward); 1058 1059 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie) 1060 { 1061 /* If it is marked obsolete, which is how we even get here, 1062 * then we have purged it from the policy bundle list and we 1063 * did that for a good reason. 1064 */ 1065 return NULL; 1066 } 1067 1068 static int stale_bundle(struct dst_entry *dst) 1069 { 1070 return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC); 1071 } 1072 1073 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev) 1074 { 1075 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) { 1076 dst->dev = &loopback_dev; 1077 dev_hold(&loopback_dev); 1078 dev_put(dev); 1079 } 1080 } 1081 EXPORT_SYMBOL(xfrm_dst_ifdown); 1082 1083 static void xfrm_link_failure(struct sk_buff *skb) 1084 { 1085 /* Impossible. Such dst must be popped before reaches point of failure. */ 1086 return; 1087 } 1088 1089 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst) 1090 { 1091 if (dst) { 1092 if (dst->obsolete) { 1093 dst_release(dst); 1094 dst = NULL; 1095 } 1096 } 1097 return dst; 1098 } 1099 1100 static void xfrm_prune_bundles(int (*func)(struct dst_entry *)) 1101 { 1102 int i; 1103 struct xfrm_policy *pol; 1104 struct dst_entry *dst, **dstp, *gc_list = NULL; 1105 1106 read_lock_bh(&xfrm_policy_lock); 1107 for (i=0; i<2*XFRM_POLICY_MAX; i++) { 1108 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) { 1109 write_lock(&pol->lock); 1110 dstp = &pol->bundles; 1111 while ((dst=*dstp) != NULL) { 1112 if (func(dst)) { 1113 *dstp = dst->next; 1114 dst->next = gc_list; 1115 gc_list = dst; 1116 } else { 1117 dstp = &dst->next; 1118 } 1119 } 1120 write_unlock(&pol->lock); 1121 } 1122 } 1123 read_unlock_bh(&xfrm_policy_lock); 1124 1125 while (gc_list) { 1126 dst = gc_list; 1127 gc_list = dst->next; 1128 dst_free(dst); 1129 } 1130 } 1131 1132 static int unused_bundle(struct dst_entry *dst) 1133 { 1134 return !atomic_read(&dst->__refcnt); 1135 } 1136 1137 static void __xfrm_garbage_collect(void) 1138 { 1139 xfrm_prune_bundles(unused_bundle); 1140 } 1141 1142 int xfrm_flush_bundles(void) 1143 { 1144 xfrm_prune_bundles(stale_bundle); 1145 return 0; 1146 } 1147 1148 static int always_true(struct dst_entry *dst) 1149 { 1150 return 1; 1151 } 1152 1153 void xfrm_flush_all_bundles(void) 1154 { 1155 xfrm_prune_bundles(always_true); 1156 } 1157 1158 void xfrm_init_pmtu(struct dst_entry *dst) 1159 { 1160 do { 1161 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 1162 u32 pmtu, route_mtu_cached; 1163 1164 pmtu = dst_mtu(dst->child); 1165 xdst->child_mtu_cached = pmtu; 1166 1167 pmtu = xfrm_state_mtu(dst->xfrm, pmtu); 1168 1169 route_mtu_cached = dst_mtu(xdst->route); 1170 xdst->route_mtu_cached = route_mtu_cached; 1171 1172 if (pmtu > route_mtu_cached) 1173 pmtu = route_mtu_cached; 1174 1175 dst->metrics[RTAX_MTU-1] = pmtu; 1176 } while ((dst = dst->next)); 1177 } 1178 1179 EXPORT_SYMBOL(xfrm_init_pmtu); 1180 1181 /* Check that the bundle accepts the flow and its components are 1182 * still valid. 1183 */ 1184 1185 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family) 1186 { 1187 struct dst_entry *dst = &first->u.dst; 1188 struct xfrm_dst *last; 1189 u32 mtu; 1190 1191 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) || 1192 (dst->dev && !netif_running(dst->dev))) 1193 return 0; 1194 1195 last = NULL; 1196 1197 do { 1198 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 1199 1200 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family)) 1201 return 0; 1202 if (dst->xfrm->km.state != XFRM_STATE_VALID) 1203 return 0; 1204 1205 mtu = dst_mtu(dst->child); 1206 if (xdst->child_mtu_cached != mtu) { 1207 last = xdst; 1208 xdst->child_mtu_cached = mtu; 1209 } 1210 1211 if (!dst_check(xdst->route, xdst->route_cookie)) 1212 return 0; 1213 mtu = dst_mtu(xdst->route); 1214 if (xdst->route_mtu_cached != mtu) { 1215 last = xdst; 1216 xdst->route_mtu_cached = mtu; 1217 } 1218 1219 dst = dst->child; 1220 } while (dst->xfrm); 1221 1222 if (likely(!last)) 1223 return 1; 1224 1225 mtu = last->child_mtu_cached; 1226 for (;;) { 1227 dst = &last->u.dst; 1228 1229 mtu = xfrm_state_mtu(dst->xfrm, mtu); 1230 if (mtu > last->route_mtu_cached) 1231 mtu = last->route_mtu_cached; 1232 dst->metrics[RTAX_MTU-1] = mtu; 1233 1234 if (last == first) 1235 break; 1236 1237 last = last->u.next; 1238 last->child_mtu_cached = mtu; 1239 } 1240 1241 return 1; 1242 } 1243 1244 EXPORT_SYMBOL(xfrm_bundle_ok); 1245 1246 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo) 1247 { 1248 int err = 0; 1249 if (unlikely(afinfo == NULL)) 1250 return -EINVAL; 1251 if (unlikely(afinfo->family >= NPROTO)) 1252 return -EAFNOSUPPORT; 1253 write_lock(&xfrm_policy_afinfo_lock); 1254 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL)) 1255 err = -ENOBUFS; 1256 else { 1257 struct dst_ops *dst_ops = afinfo->dst_ops; 1258 if (likely(dst_ops->kmem_cachep == NULL)) 1259 dst_ops->kmem_cachep = xfrm_dst_cache; 1260 if (likely(dst_ops->check == NULL)) 1261 dst_ops->check = xfrm_dst_check; 1262 if (likely(dst_ops->negative_advice == NULL)) 1263 dst_ops->negative_advice = xfrm_negative_advice; 1264 if (likely(dst_ops->link_failure == NULL)) 1265 dst_ops->link_failure = xfrm_link_failure; 1266 if (likely(afinfo->garbage_collect == NULL)) 1267 afinfo->garbage_collect = __xfrm_garbage_collect; 1268 xfrm_policy_afinfo[afinfo->family] = afinfo; 1269 } 1270 write_unlock(&xfrm_policy_afinfo_lock); 1271 return err; 1272 } 1273 EXPORT_SYMBOL(xfrm_policy_register_afinfo); 1274 1275 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo) 1276 { 1277 int err = 0; 1278 if (unlikely(afinfo == NULL)) 1279 return -EINVAL; 1280 if (unlikely(afinfo->family >= NPROTO)) 1281 return -EAFNOSUPPORT; 1282 write_lock(&xfrm_policy_afinfo_lock); 1283 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) { 1284 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo)) 1285 err = -EINVAL; 1286 else { 1287 struct dst_ops *dst_ops = afinfo->dst_ops; 1288 xfrm_policy_afinfo[afinfo->family] = NULL; 1289 dst_ops->kmem_cachep = NULL; 1290 dst_ops->check = NULL; 1291 dst_ops->negative_advice = NULL; 1292 dst_ops->link_failure = NULL; 1293 afinfo->garbage_collect = NULL; 1294 } 1295 } 1296 write_unlock(&xfrm_policy_afinfo_lock); 1297 return err; 1298 } 1299 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo); 1300 1301 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family) 1302 { 1303 struct xfrm_policy_afinfo *afinfo; 1304 if (unlikely(family >= NPROTO)) 1305 return NULL; 1306 read_lock(&xfrm_policy_afinfo_lock); 1307 afinfo = xfrm_policy_afinfo[family]; 1308 if (likely(afinfo != NULL)) 1309 read_lock(&afinfo->lock); 1310 read_unlock(&xfrm_policy_afinfo_lock); 1311 return afinfo; 1312 } 1313 1314 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo) 1315 { 1316 if (unlikely(afinfo == NULL)) 1317 return; 1318 read_unlock(&afinfo->lock); 1319 } 1320 1321 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr) 1322 { 1323 switch (event) { 1324 case NETDEV_DOWN: 1325 xfrm_flush_bundles(); 1326 } 1327 return NOTIFY_DONE; 1328 } 1329 1330 static struct notifier_block xfrm_dev_notifier = { 1331 xfrm_dev_event, 1332 NULL, 1333 0 1334 }; 1335 1336 static void __init xfrm_policy_init(void) 1337 { 1338 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache", 1339 sizeof(struct xfrm_dst), 1340 0, SLAB_HWCACHE_ALIGN, 1341 NULL, NULL); 1342 if (!xfrm_dst_cache) 1343 panic("XFRM: failed to allocate xfrm_dst_cache\n"); 1344 1345 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL); 1346 register_netdevice_notifier(&xfrm_dev_notifier); 1347 } 1348 1349 void __init xfrm_init(void) 1350 { 1351 xfrm_state_init(); 1352 xfrm_policy_init(); 1353 xfrm_input_init(); 1354 } 1355 1356