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 * Rusty Russell (C)2000 -- This code is GPL. 8 * Patrick McHardy (c) 2006-2012 9 */ 10 #include <linux/kernel.h> 11 #include <linux/netfilter.h> 12 #include <net/protocol.h> 13 #include <linux/init.h> 14 #include <linux/skbuff.h> 15 #include <linux/wait.h> 16 #include <linux/module.h> 17 #include <linux/interrupt.h> 18 #include <linux/if.h> 19 #include <linux/netdevice.h> 20 #include <linux/netfilter_ipv6.h> 21 #include <linux/inetdevice.h> 22 #include <linux/proc_fs.h> 23 #include <linux/mutex.h> 24 #include <linux/slab.h> 25 #include <linux/rcupdate.h> 26 #include <net/net_namespace.h> 27 #include <net/sock.h> 28 29 #include "nf_internals.h" 30 31 static DEFINE_MUTEX(afinfo_mutex); 32 33 const struct nf_afinfo __rcu *nf_afinfo[NFPROTO_NUMPROTO] __read_mostly; 34 EXPORT_SYMBOL(nf_afinfo); 35 const struct nf_ipv6_ops __rcu *nf_ipv6_ops __read_mostly; 36 EXPORT_SYMBOL_GPL(nf_ipv6_ops); 37 38 DEFINE_PER_CPU(bool, nf_skb_duplicated); 39 EXPORT_SYMBOL_GPL(nf_skb_duplicated); 40 41 int nf_register_afinfo(const struct nf_afinfo *afinfo) 42 { 43 mutex_lock(&afinfo_mutex); 44 RCU_INIT_POINTER(nf_afinfo[afinfo->family], afinfo); 45 mutex_unlock(&afinfo_mutex); 46 return 0; 47 } 48 EXPORT_SYMBOL_GPL(nf_register_afinfo); 49 50 void nf_unregister_afinfo(const struct nf_afinfo *afinfo) 51 { 52 mutex_lock(&afinfo_mutex); 53 RCU_INIT_POINTER(nf_afinfo[afinfo->family], NULL); 54 mutex_unlock(&afinfo_mutex); 55 synchronize_rcu(); 56 } 57 EXPORT_SYMBOL_GPL(nf_unregister_afinfo); 58 59 #ifdef HAVE_JUMP_LABEL 60 struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS]; 61 EXPORT_SYMBOL(nf_hooks_needed); 62 #endif 63 64 static DEFINE_MUTEX(nf_hook_mutex); 65 #define nf_entry_dereference(e) \ 66 rcu_dereference_protected(e, lockdep_is_held(&nf_hook_mutex)) 67 68 static struct nf_hook_entry *nf_hook_entry_head(struct net *net, 69 const struct nf_hook_ops *reg) 70 { 71 struct nf_hook_entry *hook_head = NULL; 72 73 if (reg->pf != NFPROTO_NETDEV) 74 hook_head = nf_entry_dereference(net->nf.hooks[reg->pf] 75 [reg->hooknum]); 76 else if (reg->hooknum == NF_NETDEV_INGRESS) { 77 #ifdef CONFIG_NETFILTER_INGRESS 78 if (reg->dev && dev_net(reg->dev) == net) 79 hook_head = 80 nf_entry_dereference( 81 reg->dev->nf_hooks_ingress); 82 #endif 83 } 84 return hook_head; 85 } 86 87 /* must hold nf_hook_mutex */ 88 static void nf_set_hooks_head(struct net *net, const struct nf_hook_ops *reg, 89 struct nf_hook_entry *entry) 90 { 91 switch (reg->pf) { 92 case NFPROTO_NETDEV: 93 /* We already checked in nf_register_net_hook() that this is 94 * used from ingress. 95 */ 96 rcu_assign_pointer(reg->dev->nf_hooks_ingress, entry); 97 break; 98 default: 99 rcu_assign_pointer(net->nf.hooks[reg->pf][reg->hooknum], 100 entry); 101 break; 102 } 103 } 104 105 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg) 106 { 107 struct nf_hook_entry *hooks_entry; 108 struct nf_hook_entry *entry; 109 110 if (reg->pf == NFPROTO_NETDEV && 111 (reg->hooknum != NF_NETDEV_INGRESS || 112 !reg->dev || dev_net(reg->dev) != net)) 113 return -EINVAL; 114 115 entry = kmalloc(sizeof(*entry), GFP_KERNEL); 116 if (!entry) 117 return -ENOMEM; 118 119 entry->orig_ops = reg; 120 entry->ops = *reg; 121 entry->next = NULL; 122 123 mutex_lock(&nf_hook_mutex); 124 hooks_entry = nf_hook_entry_head(net, reg); 125 126 if (hooks_entry && hooks_entry->orig_ops->priority > reg->priority) { 127 /* This is the case where we need to insert at the head */ 128 entry->next = hooks_entry; 129 hooks_entry = NULL; 130 } 131 132 while (hooks_entry && 133 reg->priority >= hooks_entry->orig_ops->priority && 134 nf_entry_dereference(hooks_entry->next)) { 135 hooks_entry = nf_entry_dereference(hooks_entry->next); 136 } 137 138 if (hooks_entry) { 139 entry->next = nf_entry_dereference(hooks_entry->next); 140 rcu_assign_pointer(hooks_entry->next, entry); 141 } else { 142 nf_set_hooks_head(net, reg, entry); 143 } 144 145 mutex_unlock(&nf_hook_mutex); 146 #ifdef CONFIG_NETFILTER_INGRESS 147 if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS) 148 net_inc_ingress_queue(); 149 #endif 150 #ifdef HAVE_JUMP_LABEL 151 static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]); 152 #endif 153 return 0; 154 } 155 EXPORT_SYMBOL(nf_register_net_hook); 156 157 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg) 158 { 159 struct nf_hook_entry *hooks_entry; 160 161 mutex_lock(&nf_hook_mutex); 162 hooks_entry = nf_hook_entry_head(net, reg); 163 if (hooks_entry && hooks_entry->orig_ops == reg) { 164 nf_set_hooks_head(net, reg, 165 nf_entry_dereference(hooks_entry->next)); 166 goto unlock; 167 } 168 while (hooks_entry && nf_entry_dereference(hooks_entry->next)) { 169 struct nf_hook_entry *next = 170 nf_entry_dereference(hooks_entry->next); 171 struct nf_hook_entry *nnext; 172 173 if (next->orig_ops != reg) { 174 hooks_entry = next; 175 continue; 176 } 177 nnext = nf_entry_dereference(next->next); 178 rcu_assign_pointer(hooks_entry->next, nnext); 179 hooks_entry = next; 180 break; 181 } 182 183 unlock: 184 mutex_unlock(&nf_hook_mutex); 185 if (!hooks_entry) { 186 WARN(1, "nf_unregister_net_hook: hook not found!\n"); 187 return; 188 } 189 #ifdef CONFIG_NETFILTER_INGRESS 190 if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS) 191 net_dec_ingress_queue(); 192 #endif 193 #ifdef HAVE_JUMP_LABEL 194 static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]); 195 #endif 196 synchronize_net(); 197 nf_queue_nf_hook_drop(net, hooks_entry); 198 /* other cpu might still process nfqueue verdict that used reg */ 199 synchronize_net(); 200 kfree(hooks_entry); 201 } 202 EXPORT_SYMBOL(nf_unregister_net_hook); 203 204 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg, 205 unsigned int n) 206 { 207 unsigned int i; 208 int err = 0; 209 210 for (i = 0; i < n; i++) { 211 err = nf_register_net_hook(net, ®[i]); 212 if (err) 213 goto err; 214 } 215 return err; 216 217 err: 218 if (i > 0) 219 nf_unregister_net_hooks(net, reg, i); 220 return err; 221 } 222 EXPORT_SYMBOL(nf_register_net_hooks); 223 224 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg, 225 unsigned int n) 226 { 227 while (n-- > 0) 228 nf_unregister_net_hook(net, ®[n]); 229 } 230 EXPORT_SYMBOL(nf_unregister_net_hooks); 231 232 static LIST_HEAD(nf_hook_list); 233 234 static int _nf_register_hook(struct nf_hook_ops *reg) 235 { 236 struct net *net, *last; 237 int ret; 238 239 for_each_net(net) { 240 ret = nf_register_net_hook(net, reg); 241 if (ret && ret != -ENOENT) 242 goto rollback; 243 } 244 list_add_tail(®->list, &nf_hook_list); 245 246 return 0; 247 rollback: 248 last = net; 249 for_each_net(net) { 250 if (net == last) 251 break; 252 nf_unregister_net_hook(net, reg); 253 } 254 return ret; 255 } 256 257 int nf_register_hook(struct nf_hook_ops *reg) 258 { 259 int ret; 260 261 rtnl_lock(); 262 ret = _nf_register_hook(reg); 263 rtnl_unlock(); 264 265 return ret; 266 } 267 EXPORT_SYMBOL(nf_register_hook); 268 269 static void _nf_unregister_hook(struct nf_hook_ops *reg) 270 { 271 struct net *net; 272 273 list_del(®->list); 274 for_each_net(net) 275 nf_unregister_net_hook(net, reg); 276 } 277 278 void nf_unregister_hook(struct nf_hook_ops *reg) 279 { 280 rtnl_lock(); 281 _nf_unregister_hook(reg); 282 rtnl_unlock(); 283 } 284 EXPORT_SYMBOL(nf_unregister_hook); 285 286 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n) 287 { 288 unsigned int i; 289 int err = 0; 290 291 for (i = 0; i < n; i++) { 292 err = nf_register_hook(®[i]); 293 if (err) 294 goto err; 295 } 296 return err; 297 298 err: 299 if (i > 0) 300 nf_unregister_hooks(reg, i); 301 return err; 302 } 303 EXPORT_SYMBOL(nf_register_hooks); 304 305 /* Caller MUST take rtnl_lock() */ 306 int _nf_register_hooks(struct nf_hook_ops *reg, unsigned int n) 307 { 308 unsigned int i; 309 int err = 0; 310 311 for (i = 0; i < n; i++) { 312 err = _nf_register_hook(®[i]); 313 if (err) 314 goto err; 315 } 316 return err; 317 318 err: 319 if (i > 0) 320 _nf_unregister_hooks(reg, i); 321 return err; 322 } 323 EXPORT_SYMBOL(_nf_register_hooks); 324 325 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n) 326 { 327 while (n-- > 0) 328 nf_unregister_hook(®[n]); 329 } 330 EXPORT_SYMBOL(nf_unregister_hooks); 331 332 /* Caller MUST take rtnl_lock */ 333 void _nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n) 334 { 335 while (n-- > 0) 336 _nf_unregister_hook(®[n]); 337 } 338 EXPORT_SYMBOL(_nf_unregister_hooks); 339 340 unsigned int nf_iterate(struct sk_buff *skb, 341 struct nf_hook_state *state, 342 struct nf_hook_entry **entryp) 343 { 344 unsigned int verdict; 345 346 /* 347 * The caller must not block between calls to this 348 * function because of risk of continuing from deleted element. 349 */ 350 while (*entryp) { 351 if (state->thresh > (*entryp)->ops.priority) { 352 *entryp = rcu_dereference((*entryp)->next); 353 continue; 354 } 355 356 /* Optimization: we don't need to hold module 357 reference here, since function can't sleep. --RR */ 358 repeat: 359 verdict = (*entryp)->ops.hook((*entryp)->ops.priv, skb, state); 360 if (verdict != NF_ACCEPT) { 361 #ifdef CONFIG_NETFILTER_DEBUG 362 if (unlikely((verdict & NF_VERDICT_MASK) 363 > NF_MAX_VERDICT)) { 364 NFDEBUG("Evil return from %p(%u).\n", 365 (*entryp)->ops.hook, state->hook); 366 *entryp = rcu_dereference((*entryp)->next); 367 continue; 368 } 369 #endif 370 if (verdict != NF_REPEAT) 371 return verdict; 372 goto repeat; 373 } 374 *entryp = rcu_dereference((*entryp)->next); 375 } 376 return NF_ACCEPT; 377 } 378 379 380 /* Returns 1 if okfn() needs to be executed by the caller, 381 * -EPERM for NF_DROP, 0 otherwise. Caller must hold rcu_read_lock. */ 382 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state) 383 { 384 struct nf_hook_entry *entry; 385 unsigned int verdict; 386 int ret = 0; 387 388 entry = rcu_dereference(state->hook_entries); 389 next_hook: 390 verdict = nf_iterate(skb, state, &entry); 391 if (verdict == NF_ACCEPT || verdict == NF_STOP) { 392 ret = 1; 393 } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) { 394 kfree_skb(skb); 395 ret = NF_DROP_GETERR(verdict); 396 if (ret == 0) 397 ret = -EPERM; 398 } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) { 399 int err; 400 401 RCU_INIT_POINTER(state->hook_entries, entry); 402 err = nf_queue(skb, state, verdict >> NF_VERDICT_QBITS); 403 if (err < 0) { 404 if (err == -ESRCH && 405 (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS)) 406 goto next_hook; 407 kfree_skb(skb); 408 } 409 } 410 return ret; 411 } 412 EXPORT_SYMBOL(nf_hook_slow); 413 414 415 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len) 416 { 417 if (writable_len > skb->len) 418 return 0; 419 420 /* Not exclusive use of packet? Must copy. */ 421 if (!skb_cloned(skb)) { 422 if (writable_len <= skb_headlen(skb)) 423 return 1; 424 } else if (skb_clone_writable(skb, writable_len)) 425 return 1; 426 427 if (writable_len <= skb_headlen(skb)) 428 writable_len = 0; 429 else 430 writable_len -= skb_headlen(skb); 431 432 return !!__pskb_pull_tail(skb, writable_len); 433 } 434 EXPORT_SYMBOL(skb_make_writable); 435 436 /* This needs to be compiled in any case to avoid dependencies between the 437 * nfnetlink_queue code and nf_conntrack. 438 */ 439 struct nfnl_ct_hook __rcu *nfnl_ct_hook __read_mostly; 440 EXPORT_SYMBOL_GPL(nfnl_ct_hook); 441 442 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 443 /* This does not belong here, but locally generated errors need it if connection 444 tracking in use: without this, connection may not be in hash table, and hence 445 manufactured ICMP or RST packets will not be associated with it. */ 446 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *) 447 __rcu __read_mostly; 448 EXPORT_SYMBOL(ip_ct_attach); 449 450 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb) 451 { 452 void (*attach)(struct sk_buff *, const struct sk_buff *); 453 454 if (skb->nfct) { 455 rcu_read_lock(); 456 attach = rcu_dereference(ip_ct_attach); 457 if (attach) 458 attach(new, skb); 459 rcu_read_unlock(); 460 } 461 } 462 EXPORT_SYMBOL(nf_ct_attach); 463 464 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly; 465 EXPORT_SYMBOL(nf_ct_destroy); 466 467 void nf_conntrack_destroy(struct nf_conntrack *nfct) 468 { 469 void (*destroy)(struct nf_conntrack *); 470 471 rcu_read_lock(); 472 destroy = rcu_dereference(nf_ct_destroy); 473 BUG_ON(destroy == NULL); 474 destroy(nfct); 475 rcu_read_unlock(); 476 } 477 EXPORT_SYMBOL(nf_conntrack_destroy); 478 479 /* Built-in default zone used e.g. by modules. */ 480 const struct nf_conntrack_zone nf_ct_zone_dflt = { 481 .id = NF_CT_DEFAULT_ZONE_ID, 482 .dir = NF_CT_DEFAULT_ZONE_DIR, 483 }; 484 EXPORT_SYMBOL_GPL(nf_ct_zone_dflt); 485 #endif /* CONFIG_NF_CONNTRACK */ 486 487 #ifdef CONFIG_NF_NAT_NEEDED 488 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *); 489 EXPORT_SYMBOL(nf_nat_decode_session_hook); 490 #endif 491 492 static int nf_register_hook_list(struct net *net) 493 { 494 struct nf_hook_ops *elem; 495 int ret; 496 497 rtnl_lock(); 498 list_for_each_entry(elem, &nf_hook_list, list) { 499 ret = nf_register_net_hook(net, elem); 500 if (ret && ret != -ENOENT) 501 goto out_undo; 502 } 503 rtnl_unlock(); 504 return 0; 505 506 out_undo: 507 list_for_each_entry_continue_reverse(elem, &nf_hook_list, list) 508 nf_unregister_net_hook(net, elem); 509 rtnl_unlock(); 510 return ret; 511 } 512 513 static void nf_unregister_hook_list(struct net *net) 514 { 515 struct nf_hook_ops *elem; 516 517 rtnl_lock(); 518 list_for_each_entry(elem, &nf_hook_list, list) 519 nf_unregister_net_hook(net, elem); 520 rtnl_unlock(); 521 } 522 523 static int __net_init netfilter_net_init(struct net *net) 524 { 525 int i, h, ret; 526 527 for (i = 0; i < ARRAY_SIZE(net->nf.hooks); i++) { 528 for (h = 0; h < NF_MAX_HOOKS; h++) 529 RCU_INIT_POINTER(net->nf.hooks[i][h], NULL); 530 } 531 532 #ifdef CONFIG_PROC_FS 533 net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter", 534 net->proc_net); 535 if (!net->nf.proc_netfilter) { 536 if (!net_eq(net, &init_net)) 537 pr_err("cannot create netfilter proc entry"); 538 539 return -ENOMEM; 540 } 541 #endif 542 ret = nf_register_hook_list(net); 543 if (ret) 544 remove_proc_entry("netfilter", net->proc_net); 545 546 return ret; 547 } 548 549 static void __net_exit netfilter_net_exit(struct net *net) 550 { 551 nf_unregister_hook_list(net); 552 remove_proc_entry("netfilter", net->proc_net); 553 } 554 555 static struct pernet_operations netfilter_net_ops = { 556 .init = netfilter_net_init, 557 .exit = netfilter_net_exit, 558 }; 559 560 int __init netfilter_init(void) 561 { 562 int ret; 563 564 ret = register_pernet_subsys(&netfilter_net_ops); 565 if (ret < 0) 566 goto err; 567 568 ret = netfilter_log_init(); 569 if (ret < 0) 570 goto err_pernet; 571 572 return 0; 573 err_pernet: 574 unregister_pernet_subsys(&netfilter_net_ops); 575 err: 576 return ret; 577 } 578