1 /* netfilter.c: look after the filters for various protocols. 2 * Heavily influenced by the old firewall.c by David Bonn and Alan Cox. 3 * 4 * Thanks to Rob `CmdrTaco' Malda for not influencing this code in any 5 * way. 6 * 7 * This code is GPL. 8 */ 9 #include <linux/kernel.h> 10 #include <linux/netfilter.h> 11 #include <net/protocol.h> 12 #include <linux/init.h> 13 #include <linux/skbuff.h> 14 #include <linux/wait.h> 15 #include <linux/module.h> 16 #include <linux/interrupt.h> 17 #include <linux/if.h> 18 #include <linux/netdevice.h> 19 #include <linux/netfilter_ipv6.h> 20 #include <linux/inetdevice.h> 21 #include <linux/proc_fs.h> 22 #include <linux/mutex.h> 23 #include <linux/mm.h> 24 #include <linux/rcupdate.h> 25 #include <net/net_namespace.h> 26 #include <net/netfilter/nf_queue.h> 27 #include <net/sock.h> 28 29 #include "nf_internals.h" 30 31 const struct nf_ipv6_ops __rcu *nf_ipv6_ops __read_mostly; 32 EXPORT_SYMBOL_GPL(nf_ipv6_ops); 33 34 DEFINE_PER_CPU(bool, nf_skb_duplicated); 35 EXPORT_SYMBOL_GPL(nf_skb_duplicated); 36 37 #ifdef CONFIG_JUMP_LABEL 38 struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS]; 39 EXPORT_SYMBOL(nf_hooks_needed); 40 #endif 41 42 static DEFINE_MUTEX(nf_hook_mutex); 43 44 /* max hooks per family/hooknum */ 45 #define MAX_HOOK_COUNT 1024 46 47 #define nf_entry_dereference(e) \ 48 rcu_dereference_protected(e, lockdep_is_held(&nf_hook_mutex)) 49 50 static struct nf_hook_entries *allocate_hook_entries_size(u16 num) 51 { 52 struct nf_hook_entries *e; 53 size_t alloc = sizeof(*e) + 54 sizeof(struct nf_hook_entry) * num + 55 sizeof(struct nf_hook_ops *) * num + 56 sizeof(struct nf_hook_entries_rcu_head); 57 58 if (num == 0) 59 return NULL; 60 61 e = kvzalloc(alloc, GFP_KERNEL); 62 if (e) 63 e->num_hook_entries = num; 64 return e; 65 } 66 67 static void __nf_hook_entries_free(struct rcu_head *h) 68 { 69 struct nf_hook_entries_rcu_head *head; 70 71 head = container_of(h, struct nf_hook_entries_rcu_head, head); 72 kvfree(head->allocation); 73 } 74 75 static void nf_hook_entries_free(struct nf_hook_entries *e) 76 { 77 struct nf_hook_entries_rcu_head *head; 78 struct nf_hook_ops **ops; 79 unsigned int num; 80 81 if (!e) 82 return; 83 84 num = e->num_hook_entries; 85 ops = nf_hook_entries_get_hook_ops(e); 86 head = (void *)&ops[num]; 87 head->allocation = e; 88 call_rcu(&head->head, __nf_hook_entries_free); 89 } 90 91 static unsigned int accept_all(void *priv, 92 struct sk_buff *skb, 93 const struct nf_hook_state *state) 94 { 95 return NF_ACCEPT; /* ACCEPT makes nf_hook_slow call next hook */ 96 } 97 98 static const struct nf_hook_ops dummy_ops = { 99 .hook = accept_all, 100 .priority = INT_MIN, 101 }; 102 103 static struct nf_hook_entries * 104 nf_hook_entries_grow(const struct nf_hook_entries *old, 105 const struct nf_hook_ops *reg) 106 { 107 unsigned int i, alloc_entries, nhooks, old_entries; 108 struct nf_hook_ops **orig_ops = NULL; 109 struct nf_hook_ops **new_ops; 110 struct nf_hook_entries *new; 111 bool inserted = false; 112 113 alloc_entries = 1; 114 old_entries = old ? old->num_hook_entries : 0; 115 116 if (old) { 117 orig_ops = nf_hook_entries_get_hook_ops(old); 118 119 for (i = 0; i < old_entries; i++) { 120 if (orig_ops[i] != &dummy_ops) 121 alloc_entries++; 122 } 123 } 124 125 if (alloc_entries > MAX_HOOK_COUNT) 126 return ERR_PTR(-E2BIG); 127 128 new = allocate_hook_entries_size(alloc_entries); 129 if (!new) 130 return ERR_PTR(-ENOMEM); 131 132 new_ops = nf_hook_entries_get_hook_ops(new); 133 134 i = 0; 135 nhooks = 0; 136 while (i < old_entries) { 137 if (orig_ops[i] == &dummy_ops) { 138 ++i; 139 continue; 140 } 141 142 if (inserted || reg->priority > orig_ops[i]->priority) { 143 new_ops[nhooks] = (void *)orig_ops[i]; 144 new->hooks[nhooks] = old->hooks[i]; 145 i++; 146 } else { 147 new_ops[nhooks] = (void *)reg; 148 new->hooks[nhooks].hook = reg->hook; 149 new->hooks[nhooks].priv = reg->priv; 150 inserted = true; 151 } 152 nhooks++; 153 } 154 155 if (!inserted) { 156 new_ops[nhooks] = (void *)reg; 157 new->hooks[nhooks].hook = reg->hook; 158 new->hooks[nhooks].priv = reg->priv; 159 } 160 161 return new; 162 } 163 164 static void hooks_validate(const struct nf_hook_entries *hooks) 165 { 166 #ifdef CONFIG_DEBUG_MISC 167 struct nf_hook_ops **orig_ops; 168 int prio = INT_MIN; 169 size_t i = 0; 170 171 orig_ops = nf_hook_entries_get_hook_ops(hooks); 172 173 for (i = 0; i < hooks->num_hook_entries; i++) { 174 if (orig_ops[i] == &dummy_ops) 175 continue; 176 177 WARN_ON(orig_ops[i]->priority < prio); 178 179 if (orig_ops[i]->priority > prio) 180 prio = orig_ops[i]->priority; 181 } 182 #endif 183 } 184 185 int nf_hook_entries_insert_raw(struct nf_hook_entries __rcu **pp, 186 const struct nf_hook_ops *reg) 187 { 188 struct nf_hook_entries *new_hooks; 189 struct nf_hook_entries *p; 190 191 p = rcu_dereference_raw(*pp); 192 new_hooks = nf_hook_entries_grow(p, reg); 193 if (IS_ERR(new_hooks)) 194 return PTR_ERR(new_hooks); 195 196 hooks_validate(new_hooks); 197 198 rcu_assign_pointer(*pp, new_hooks); 199 200 BUG_ON(p == new_hooks); 201 nf_hook_entries_free(p); 202 return 0; 203 } 204 EXPORT_SYMBOL_GPL(nf_hook_entries_insert_raw); 205 206 /* 207 * __nf_hook_entries_try_shrink - try to shrink hook array 208 * 209 * @old -- current hook blob at @pp 210 * @pp -- location of hook blob 211 * 212 * Hook unregistration must always succeed, so to-be-removed hooks 213 * are replaced by a dummy one that will just move to next hook. 214 * 215 * This counts the current dummy hooks, attempts to allocate new blob, 216 * copies the live hooks, then replaces and discards old one. 217 * 218 * return values: 219 * 220 * Returns address to free, or NULL. 221 */ 222 static void *__nf_hook_entries_try_shrink(struct nf_hook_entries *old, 223 struct nf_hook_entries __rcu **pp) 224 { 225 unsigned int i, j, skip = 0, hook_entries; 226 struct nf_hook_entries *new = NULL; 227 struct nf_hook_ops **orig_ops; 228 struct nf_hook_ops **new_ops; 229 230 if (WARN_ON_ONCE(!old)) 231 return NULL; 232 233 orig_ops = nf_hook_entries_get_hook_ops(old); 234 for (i = 0; i < old->num_hook_entries; i++) { 235 if (orig_ops[i] == &dummy_ops) 236 skip++; 237 } 238 239 /* if skip == hook_entries all hooks have been removed */ 240 hook_entries = old->num_hook_entries; 241 if (skip == hook_entries) 242 goto out_assign; 243 244 if (skip == 0) 245 return NULL; 246 247 hook_entries -= skip; 248 new = allocate_hook_entries_size(hook_entries); 249 if (!new) 250 return NULL; 251 252 new_ops = nf_hook_entries_get_hook_ops(new); 253 for (i = 0, j = 0; i < old->num_hook_entries; i++) { 254 if (orig_ops[i] == &dummy_ops) 255 continue; 256 new->hooks[j] = old->hooks[i]; 257 new_ops[j] = (void *)orig_ops[i]; 258 j++; 259 } 260 hooks_validate(new); 261 out_assign: 262 rcu_assign_pointer(*pp, new); 263 return old; 264 } 265 266 static struct nf_hook_entries __rcu ** 267 nf_hook_entry_head(struct net *net, int pf, unsigned int hooknum, 268 struct net_device *dev) 269 { 270 switch (pf) { 271 case NFPROTO_NETDEV: 272 break; 273 #ifdef CONFIG_NETFILTER_FAMILY_ARP 274 case NFPROTO_ARP: 275 if (WARN_ON_ONCE(ARRAY_SIZE(net->nf.hooks_arp) <= hooknum)) 276 return NULL; 277 return net->nf.hooks_arp + hooknum; 278 #endif 279 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE 280 case NFPROTO_BRIDGE: 281 if (WARN_ON_ONCE(ARRAY_SIZE(net->nf.hooks_bridge) <= hooknum)) 282 return NULL; 283 return net->nf.hooks_bridge + hooknum; 284 #endif 285 case NFPROTO_IPV4: 286 if (WARN_ON_ONCE(ARRAY_SIZE(net->nf.hooks_ipv4) <= hooknum)) 287 return NULL; 288 return net->nf.hooks_ipv4 + hooknum; 289 case NFPROTO_IPV6: 290 if (WARN_ON_ONCE(ARRAY_SIZE(net->nf.hooks_ipv6) <= hooknum)) 291 return NULL; 292 return net->nf.hooks_ipv6 + hooknum; 293 #if IS_ENABLED(CONFIG_DECNET) 294 case NFPROTO_DECNET: 295 if (WARN_ON_ONCE(ARRAY_SIZE(net->nf.hooks_decnet) <= hooknum)) 296 return NULL; 297 return net->nf.hooks_decnet + hooknum; 298 #endif 299 default: 300 WARN_ON_ONCE(1); 301 return NULL; 302 } 303 304 #ifdef CONFIG_NETFILTER_INGRESS 305 if (hooknum == NF_NETDEV_INGRESS) { 306 if (dev && dev_net(dev) == net) 307 return &dev->nf_hooks_ingress; 308 } 309 #endif 310 WARN_ON_ONCE(1); 311 return NULL; 312 } 313 314 static int __nf_register_net_hook(struct net *net, int pf, 315 const struct nf_hook_ops *reg) 316 { 317 struct nf_hook_entries *p, *new_hooks; 318 struct nf_hook_entries __rcu **pp; 319 320 if (pf == NFPROTO_NETDEV) { 321 #ifndef CONFIG_NETFILTER_INGRESS 322 if (reg->hooknum == NF_NETDEV_INGRESS) 323 return -EOPNOTSUPP; 324 #endif 325 if (reg->hooknum != NF_NETDEV_INGRESS || 326 !reg->dev || dev_net(reg->dev) != net) 327 return -EINVAL; 328 } 329 330 pp = nf_hook_entry_head(net, pf, reg->hooknum, reg->dev); 331 if (!pp) 332 return -EINVAL; 333 334 mutex_lock(&nf_hook_mutex); 335 336 p = nf_entry_dereference(*pp); 337 new_hooks = nf_hook_entries_grow(p, reg); 338 339 if (!IS_ERR(new_hooks)) 340 rcu_assign_pointer(*pp, new_hooks); 341 342 mutex_unlock(&nf_hook_mutex); 343 if (IS_ERR(new_hooks)) 344 return PTR_ERR(new_hooks); 345 346 hooks_validate(new_hooks); 347 #ifdef CONFIG_NETFILTER_INGRESS 348 if (pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS) 349 net_inc_ingress_queue(); 350 #endif 351 #ifdef CONFIG_JUMP_LABEL 352 static_key_slow_inc(&nf_hooks_needed[pf][reg->hooknum]); 353 #endif 354 BUG_ON(p == new_hooks); 355 nf_hook_entries_free(p); 356 return 0; 357 } 358 359 /* 360 * nf_remove_net_hook - remove a hook from blob 361 * 362 * @oldp: current address of hook blob 363 * @unreg: hook to unregister 364 * 365 * This cannot fail, hook unregistration must always succeed. 366 * Therefore replace the to-be-removed hook with a dummy hook. 367 */ 368 static bool nf_remove_net_hook(struct nf_hook_entries *old, 369 const struct nf_hook_ops *unreg) 370 { 371 struct nf_hook_ops **orig_ops; 372 unsigned int i; 373 374 orig_ops = nf_hook_entries_get_hook_ops(old); 375 for (i = 0; i < old->num_hook_entries; i++) { 376 if (orig_ops[i] != unreg) 377 continue; 378 WRITE_ONCE(old->hooks[i].hook, accept_all); 379 WRITE_ONCE(orig_ops[i], &dummy_ops); 380 return true; 381 } 382 383 return false; 384 } 385 386 static void __nf_unregister_net_hook(struct net *net, int pf, 387 const struct nf_hook_ops *reg) 388 { 389 struct nf_hook_entries __rcu **pp; 390 struct nf_hook_entries *p; 391 392 pp = nf_hook_entry_head(net, pf, reg->hooknum, reg->dev); 393 if (!pp) 394 return; 395 396 mutex_lock(&nf_hook_mutex); 397 398 p = nf_entry_dereference(*pp); 399 if (WARN_ON_ONCE(!p)) { 400 mutex_unlock(&nf_hook_mutex); 401 return; 402 } 403 404 if (nf_remove_net_hook(p, reg)) { 405 #ifdef CONFIG_NETFILTER_INGRESS 406 if (pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS) 407 net_dec_ingress_queue(); 408 #endif 409 #ifdef CONFIG_JUMP_LABEL 410 static_key_slow_dec(&nf_hooks_needed[pf][reg->hooknum]); 411 #endif 412 } else { 413 WARN_ONCE(1, "hook not found, pf %d num %d", pf, reg->hooknum); 414 } 415 416 p = __nf_hook_entries_try_shrink(p, pp); 417 mutex_unlock(&nf_hook_mutex); 418 if (!p) 419 return; 420 421 nf_queue_nf_hook_drop(net); 422 nf_hook_entries_free(p); 423 } 424 425 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg) 426 { 427 if (reg->pf == NFPROTO_INET) { 428 __nf_unregister_net_hook(net, NFPROTO_IPV4, reg); 429 __nf_unregister_net_hook(net, NFPROTO_IPV6, reg); 430 } else { 431 __nf_unregister_net_hook(net, reg->pf, reg); 432 } 433 } 434 EXPORT_SYMBOL(nf_unregister_net_hook); 435 436 void nf_hook_entries_delete_raw(struct nf_hook_entries __rcu **pp, 437 const struct nf_hook_ops *reg) 438 { 439 struct nf_hook_entries *p; 440 441 p = rcu_dereference_raw(*pp); 442 if (nf_remove_net_hook(p, reg)) { 443 p = __nf_hook_entries_try_shrink(p, pp); 444 nf_hook_entries_free(p); 445 } 446 } 447 EXPORT_SYMBOL_GPL(nf_hook_entries_delete_raw); 448 449 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg) 450 { 451 int err; 452 453 if (reg->pf == NFPROTO_INET) { 454 err = __nf_register_net_hook(net, NFPROTO_IPV4, reg); 455 if (err < 0) 456 return err; 457 458 err = __nf_register_net_hook(net, NFPROTO_IPV6, reg); 459 if (err < 0) { 460 __nf_unregister_net_hook(net, NFPROTO_IPV4, reg); 461 return err; 462 } 463 } else { 464 err = __nf_register_net_hook(net, reg->pf, reg); 465 if (err < 0) 466 return err; 467 } 468 469 return 0; 470 } 471 EXPORT_SYMBOL(nf_register_net_hook); 472 473 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg, 474 unsigned int n) 475 { 476 unsigned int i; 477 int err = 0; 478 479 for (i = 0; i < n; i++) { 480 err = nf_register_net_hook(net, ®[i]); 481 if (err) 482 goto err; 483 } 484 return err; 485 486 err: 487 if (i > 0) 488 nf_unregister_net_hooks(net, reg, i); 489 return err; 490 } 491 EXPORT_SYMBOL(nf_register_net_hooks); 492 493 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg, 494 unsigned int hookcount) 495 { 496 unsigned int i; 497 498 for (i = 0; i < hookcount; i++) 499 nf_unregister_net_hook(net, ®[i]); 500 } 501 EXPORT_SYMBOL(nf_unregister_net_hooks); 502 503 /* Returns 1 if okfn() needs to be executed by the caller, 504 * -EPERM for NF_DROP, 0 otherwise. Caller must hold rcu_read_lock. */ 505 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state, 506 const struct nf_hook_entries *e, unsigned int s) 507 { 508 unsigned int verdict; 509 int ret; 510 511 for (; s < e->num_hook_entries; s++) { 512 verdict = nf_hook_entry_hookfn(&e->hooks[s], skb, state); 513 switch (verdict & NF_VERDICT_MASK) { 514 case NF_ACCEPT: 515 break; 516 case NF_DROP: 517 kfree_skb(skb); 518 ret = NF_DROP_GETERR(verdict); 519 if (ret == 0) 520 ret = -EPERM; 521 return ret; 522 case NF_QUEUE: 523 ret = nf_queue(skb, state, s, verdict); 524 if (ret == 1) 525 continue; 526 return ret; 527 default: 528 /* Implicit handling for NF_STOLEN, as well as any other 529 * non conventional verdicts. 530 */ 531 return 0; 532 } 533 } 534 535 return 1; 536 } 537 EXPORT_SYMBOL(nf_hook_slow); 538 539 void nf_hook_slow_list(struct list_head *head, struct nf_hook_state *state, 540 const struct nf_hook_entries *e) 541 { 542 struct sk_buff *skb, *next; 543 struct list_head sublist; 544 int ret; 545 546 INIT_LIST_HEAD(&sublist); 547 548 list_for_each_entry_safe(skb, next, head, list) { 549 skb_list_del_init(skb); 550 ret = nf_hook_slow(skb, state, e, 0); 551 if (ret == 1) 552 list_add_tail(&skb->list, &sublist); 553 } 554 /* Put passed packets back on main list */ 555 list_splice(&sublist, head); 556 } 557 EXPORT_SYMBOL(nf_hook_slow_list); 558 559 /* This needs to be compiled in any case to avoid dependencies between the 560 * nfnetlink_queue code and nf_conntrack. 561 */ 562 struct nfnl_ct_hook __rcu *nfnl_ct_hook __read_mostly; 563 EXPORT_SYMBOL_GPL(nfnl_ct_hook); 564 565 struct nf_ct_hook __rcu *nf_ct_hook __read_mostly; 566 EXPORT_SYMBOL_GPL(nf_ct_hook); 567 568 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 569 /* This does not belong here, but locally generated errors need it if connection 570 tracking in use: without this, connection may not be in hash table, and hence 571 manufactured ICMP or RST packets will not be associated with it. */ 572 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *) 573 __rcu __read_mostly; 574 EXPORT_SYMBOL(ip_ct_attach); 575 576 struct nf_nat_hook __rcu *nf_nat_hook __read_mostly; 577 EXPORT_SYMBOL_GPL(nf_nat_hook); 578 579 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb) 580 { 581 void (*attach)(struct sk_buff *, const struct sk_buff *); 582 583 if (skb->_nfct) { 584 rcu_read_lock(); 585 attach = rcu_dereference(ip_ct_attach); 586 if (attach) 587 attach(new, skb); 588 rcu_read_unlock(); 589 } 590 } 591 EXPORT_SYMBOL(nf_ct_attach); 592 593 void nf_conntrack_destroy(struct nf_conntrack *nfct) 594 { 595 struct nf_ct_hook *ct_hook; 596 597 rcu_read_lock(); 598 ct_hook = rcu_dereference(nf_ct_hook); 599 BUG_ON(ct_hook == NULL); 600 ct_hook->destroy(nfct); 601 rcu_read_unlock(); 602 } 603 EXPORT_SYMBOL(nf_conntrack_destroy); 604 605 bool nf_ct_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple, 606 const struct sk_buff *skb) 607 { 608 struct nf_ct_hook *ct_hook; 609 bool ret = false; 610 611 rcu_read_lock(); 612 ct_hook = rcu_dereference(nf_ct_hook); 613 if (ct_hook) 614 ret = ct_hook->get_tuple_skb(dst_tuple, skb); 615 rcu_read_unlock(); 616 return ret; 617 } 618 EXPORT_SYMBOL(nf_ct_get_tuple_skb); 619 620 /* Built-in default zone used e.g. by modules. */ 621 const struct nf_conntrack_zone nf_ct_zone_dflt = { 622 .id = NF_CT_DEFAULT_ZONE_ID, 623 .dir = NF_CT_DEFAULT_ZONE_DIR, 624 }; 625 EXPORT_SYMBOL_GPL(nf_ct_zone_dflt); 626 #endif /* CONFIG_NF_CONNTRACK */ 627 628 static void __net_init 629 __netfilter_net_init(struct nf_hook_entries __rcu **e, int max) 630 { 631 int h; 632 633 for (h = 0; h < max; h++) 634 RCU_INIT_POINTER(e[h], NULL); 635 } 636 637 static int __net_init netfilter_net_init(struct net *net) 638 { 639 __netfilter_net_init(net->nf.hooks_ipv4, ARRAY_SIZE(net->nf.hooks_ipv4)); 640 __netfilter_net_init(net->nf.hooks_ipv6, ARRAY_SIZE(net->nf.hooks_ipv6)); 641 #ifdef CONFIG_NETFILTER_FAMILY_ARP 642 __netfilter_net_init(net->nf.hooks_arp, ARRAY_SIZE(net->nf.hooks_arp)); 643 #endif 644 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE 645 __netfilter_net_init(net->nf.hooks_bridge, ARRAY_SIZE(net->nf.hooks_bridge)); 646 #endif 647 #if IS_ENABLED(CONFIG_DECNET) 648 __netfilter_net_init(net->nf.hooks_decnet, ARRAY_SIZE(net->nf.hooks_decnet)); 649 #endif 650 651 #ifdef CONFIG_PROC_FS 652 net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter", 653 net->proc_net); 654 if (!net->nf.proc_netfilter) { 655 if (!net_eq(net, &init_net)) 656 pr_err("cannot create netfilter proc entry"); 657 658 return -ENOMEM; 659 } 660 #endif 661 662 return 0; 663 } 664 665 static void __net_exit netfilter_net_exit(struct net *net) 666 { 667 remove_proc_entry("netfilter", net->proc_net); 668 } 669 670 static struct pernet_operations netfilter_net_ops = { 671 .init = netfilter_net_init, 672 .exit = netfilter_net_exit, 673 }; 674 675 int __init netfilter_init(void) 676 { 677 int ret; 678 679 ret = register_pernet_subsys(&netfilter_net_ops); 680 if (ret < 0) 681 goto err; 682 683 ret = netfilter_log_init(); 684 if (ret < 0) 685 goto err_pernet; 686 687 return 0; 688 err_pernet: 689 unregister_pernet_subsys(&netfilter_net_ops); 690 err: 691 return ret; 692 } 693