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