net/netfilter/core.c

   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/mm.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
  66 /* max hooks per family/hooknum */
  67 #define MAX_HOOK_COUNT          1024
  68
  69 #define nf_entry_dereference(e) \
  70         rcu_dereference_protected(e, lockdep_is_held(&nf_hook_mutex))
  71
  72 static struct nf_hook_entries *allocate_hook_entries_size(u16 num)
  73 {
  74         struct nf_hook_entries *e;
  75         size_t alloc = sizeof(*e) +
  76                        sizeof(struct nf_hook_entry) * num +
  77                        sizeof(struct nf_hook_ops *) * num;
  78
  79         if (num == 0)
  80                 return NULL;
  81
  82         e = kvzalloc(alloc, GFP_KERNEL);
  83         if (e)
  84                 e->num_hook_entries = num;
  85         return e;
  86 }
  87
  88 static unsigned int accept_all(void *priv,
  89                                struct sk_buff *skb,
  90                                const struct nf_hook_state *state)
  91 {
  92         return NF_ACCEPT; /* ACCEPT makes nf_hook_slow call next hook */
  93 }
  94
  95 static const struct nf_hook_ops dummy_ops = {
  96         .hook = accept_all,
  97         .priority = INT_MIN,
  98 };
  99
 100 static struct nf_hook_entries *
 101 nf_hook_entries_grow(const struct nf_hook_entries *old,
 102                      const struct nf_hook_ops *reg)
 103 {
 104         unsigned int i, alloc_entries, nhooks, old_entries;
 105         struct nf_hook_ops **orig_ops = NULL;
 106         struct nf_hook_ops **new_ops;
 107         struct nf_hook_entries *new;
 108         bool inserted = false;
 109
 110         alloc_entries = 1;
 111         old_entries = old ? old->num_hook_entries : 0;
 112
 113         if (old) {
 114                 orig_ops = nf_hook_entries_get_hook_ops(old);
 115
 116                 for (i = 0; i < old_entries; i++) {
 117                         if (orig_ops[i] != &dummy_ops)
 118                                 alloc_entries++;
 119                 }
 120         }
 121
 122         if (alloc_entries > MAX_HOOK_COUNT)
 123                 return ERR_PTR(-E2BIG);
 124
 125         new = allocate_hook_entries_size(alloc_entries);
 126         if (!new)
 127                 return ERR_PTR(-ENOMEM);
 128
 129         new_ops = nf_hook_entries_get_hook_ops(new);
 130
 131         i = 0;
 132         nhooks = 0;
 133         while (i < old_entries) {
 134                 if (orig_ops[i] == &dummy_ops) {
 135                         ++i;
 136                         continue;
 137                 }
 138                 if (inserted || reg->priority > orig_ops[i]->priority) {
 139                         new_ops[nhooks] = (void *)orig_ops[i];
 140                         new->hooks[nhooks] = old->hooks[i];
 141                         i++;
 142                 } else {
 143                         new_ops[nhooks] = (void *)reg;
 144                         new->hooks[nhooks].hook = reg->hook;
 145                         new->hooks[nhooks].priv = reg->priv;
 146                         inserted = true;
 147                 }
 148                 nhooks++;
 149         }
 150
 151         if (!inserted) {
 152                 new_ops[nhooks] = (void *)reg;
 153                 new->hooks[nhooks].hook = reg->hook;
 154                 new->hooks[nhooks].priv = reg->priv;
 155         }
 156
 157         return new;
 158 }
 159
 160 static void hooks_validate(const struct nf_hook_entries *hooks)
 161 {
 162 #ifdef CONFIG_DEBUG_KERNEL
 163         struct nf_hook_ops **orig_ops;
 164         int prio = INT_MIN;
 165         size_t i = 0;
 166
 167         orig_ops = nf_hook_entries_get_hook_ops(hooks);
 168
 169         for (i = 0; i < hooks->num_hook_entries; i++) {
 170                 if (orig_ops[i] == &dummy_ops)
 171                         continue;
 172
 173                 WARN_ON(orig_ops[i]->priority < prio);
 174
 175                 if (orig_ops[i]->priority > prio)
 176                         prio = orig_ops[i]->priority;
 177         }
 178 #endif
 179 }
 180
 181 /*
 182  * __nf_hook_entries_try_shrink - try to shrink hook array
 183  *
 184  * @pp -- location of hook blob
 185  *
 186  * Hook unregistration must always succeed, so to-be-removed hooks
 187  * are replaced by a dummy one that will just move to next hook.
 188  *
 189  * This counts the current dummy hooks, attempts to allocate new blob,
 190  * copies the live hooks, then replaces and discards old one.
 191  *
 192  * return values:
 193  *
 194  * Returns address to free, or NULL.
 195  */
 196 static void *__nf_hook_entries_try_shrink(struct nf_hook_entries __rcu **pp)
 197 {
 198         struct nf_hook_entries *old, *new = NULL;
 199         unsigned int i, j, skip = 0, hook_entries;
 200         struct nf_hook_ops **orig_ops;
 201         struct nf_hook_ops **new_ops;
 202
 203         old = nf_entry_dereference(*pp);
 204         if (WARN_ON_ONCE(!old))
 205                 return NULL;
 206
 207         orig_ops = nf_hook_entries_get_hook_ops(old);
 208         for (i = 0; i < old->num_hook_entries; i++) {
 209                 if (orig_ops[i] == &dummy_ops)
 210                         skip++;
 211         }
 212
 213         /* if skip == hook_entries all hooks have been removed */
 214         hook_entries = old->num_hook_entries;
 215         if (skip == hook_entries)
 216                 goto out_assign;
 217
 218         if (skip == 0)
 219                 return NULL;
 220
 221         hook_entries -= skip;
 222         new = allocate_hook_entries_size(hook_entries);
 223         if (!new)
 224                 return NULL;
 225
 226         new_ops = nf_hook_entries_get_hook_ops(new);
 227         for (i = 0, j = 0; i < old->num_hook_entries; i++) {
 228                 if (orig_ops[i] == &dummy_ops)
 229                         continue;
 230                 new->hooks[j] = old->hooks[i];
 231                 new_ops[j] = (void *)orig_ops[i];
 232                 j++;
 233         }
 234         hooks_validate(new);
 235 out_assign:
 236         rcu_assign_pointer(*pp, new);
 237         return old;
 238 }
 239
 240 static struct nf_hook_entries __rcu **nf_hook_entry_head(struct net *net, const struct nf_hook_ops *reg)
 241 {
 242         if (reg->pf != NFPROTO_NETDEV)
 243                 return net->nf.hooks[reg->pf]+reg->hooknum;
 244
 245 #ifdef CONFIG_NETFILTER_INGRESS
 246         if (reg->hooknum == NF_NETDEV_INGRESS) {
 247                 if (reg->dev && dev_net(reg->dev) == net)
 248                         return &reg->dev->nf_hooks_ingress;
 249         }
 250 #endif
 251         WARN_ON_ONCE(1);
 252         return NULL;
 253 }
 254
 255 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
 256 {
 257         struct nf_hook_entries *p, *new_hooks;
 258         struct nf_hook_entries __rcu **pp;
 259
 260         if (reg->pf == NFPROTO_NETDEV) {
 261 #ifndef CONFIG_NETFILTER_INGRESS
 262                 if (reg->hooknum == NF_NETDEV_INGRESS)
 263                         return -EOPNOTSUPP;
 264 #endif
 265                 if (reg->hooknum != NF_NETDEV_INGRESS ||
 266                     !reg->dev || dev_net(reg->dev) != net)
 267                         return -EINVAL;
 268         }
 269
 270         pp = nf_hook_entry_head(net, reg);
 271         if (!pp)
 272                 return -EINVAL;
 273
 274         mutex_lock(&nf_hook_mutex);
 275
 276         p = nf_entry_dereference(*pp);
 277         new_hooks = nf_hook_entries_grow(p, reg);
 278
 279         if (!IS_ERR(new_hooks))
 280                 rcu_assign_pointer(*pp, new_hooks);
 281
 282         mutex_unlock(&nf_hook_mutex);
 283         if (IS_ERR(new_hooks))
 284                 return PTR_ERR(new_hooks);
 285
 286         hooks_validate(new_hooks);
 287 #ifdef CONFIG_NETFILTER_INGRESS
 288         if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 289                 net_inc_ingress_queue();
 290 #endif
 291 #ifdef HAVE_JUMP_LABEL
 292         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
 293 #endif
 294         synchronize_net();
 295         BUG_ON(p == new_hooks);
 296         kvfree(p);
 297         return 0;
 298 }
 299 EXPORT_SYMBOL(nf_register_net_hook);
 300
 301 /*
 302  * __nf_unregister_net_hook - remove a hook from blob
 303  *
 304  * @oldp: current address of hook blob
 305  * @unreg: hook to unregister
 306  *
 307  * This cannot fail, hook unregistration must always succeed.
 308  * Therefore replace the to-be-removed hook with a dummy hook.
 309  */
 310 static void __nf_unregister_net_hook(struct nf_hook_entries *old,
 311                                      const struct nf_hook_ops *unreg)
 312 {
 313         struct nf_hook_ops **orig_ops;
 314         bool found = false;
 315         unsigned int i;
 316
 317         orig_ops = nf_hook_entries_get_hook_ops(old);
 318         for (i = 0; i < old->num_hook_entries; i++) {
 319                 if (orig_ops[i] != unreg)
 320                         continue;
 321                 WRITE_ONCE(old->hooks[i].hook, accept_all);
 322                 WRITE_ONCE(orig_ops[i], &dummy_ops);
 323                 found = true;
 324                 break;
 325         }
 326
 327         if (found) {
 328 #ifdef CONFIG_NETFILTER_INGRESS
 329                 if (unreg->pf == NFPROTO_NETDEV && unreg->hooknum == NF_NETDEV_INGRESS)
 330                         net_dec_ingress_queue();
 331 #endif
 332 #ifdef HAVE_JUMP_LABEL
 333                 static_key_slow_dec(&nf_hooks_needed[unreg->pf][unreg->hooknum]);
 334 #endif
 335         } else {
 336                 WARN_ONCE(1, "hook not found, pf %d num %d", unreg->pf, unreg->hooknum);
 337         }
 338 }
 339
 340 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 341 {
 342         struct nf_hook_entries __rcu **pp;
 343         struct nf_hook_entries *p;
 344         unsigned int nfq;
 345
 346         pp = nf_hook_entry_head(net, reg);
 347         if (!pp)
 348                 return;
 349
 350         mutex_lock(&nf_hook_mutex);
 351
 352         p = nf_entry_dereference(*pp);
 353         if (WARN_ON_ONCE(!p)) {
 354                 mutex_unlock(&nf_hook_mutex);
 355                 return;
 356         }
 357
 358         __nf_unregister_net_hook(p, reg);
 359
 360         p = __nf_hook_entries_try_shrink(pp);
 361         mutex_unlock(&nf_hook_mutex);
 362         if (!p)
 363                 return;
 364
 365         synchronize_net();
 366
 367         /* other cpu might still process nfqueue verdict that used reg */
 368         nfq = nf_queue_nf_hook_drop(net);
 369         if (nfq)
 370                 synchronize_net();
 371         kvfree(p);
 372 }
 373 EXPORT_SYMBOL(nf_unregister_net_hook);
 374
 375 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 376                           unsigned int n)
 377 {
 378         unsigned int i;
 379         int err = 0;
 380
 381         for (i = 0; i < n; i++) {
 382                 err = nf_register_net_hook(net, &reg[i]);
 383                 if (err)
 384                         goto err;
 385         }
 386         return err;
 387
 388 err:
 389         if (i > 0)
 390                 nf_unregister_net_hooks(net, reg, i);
 391         return err;
 392 }
 393 EXPORT_SYMBOL(nf_register_net_hooks);
 394
 395 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 396                              unsigned int hookcount)
 397 {
 398         struct nf_hook_entries *to_free[16], *p;
 399         struct nf_hook_entries __rcu **pp;
 400         unsigned int i, j, n;
 401
 402         mutex_lock(&nf_hook_mutex);
 403         for (i = 0; i < hookcount; i++) {
 404                 pp = nf_hook_entry_head(net, &reg[i]);
 405                 if (!pp)
 406                         continue;
 407
 408                 p = nf_entry_dereference(*pp);
 409                 if (WARN_ON_ONCE(!p))
 410                         continue;
 411                 __nf_unregister_net_hook(p, &reg[i]);
 412         }
 413         mutex_unlock(&nf_hook_mutex);
 414
 415         do {
 416                 n = min_t(unsigned int, hookcount, ARRAY_SIZE(to_free));
 417
 418                 mutex_lock(&nf_hook_mutex);
 419
 420                 for (i = 0, j = 0; i < hookcount && j < n; i++) {
 421                         pp = nf_hook_entry_head(net, &reg[i]);
 422                         if (!pp)
 423                                 continue;
 424
 425                         p = nf_entry_dereference(*pp);
 426                         if (!p)
 427                                 continue;
 428
 429                         to_free[j] = __nf_hook_entries_try_shrink(pp);
 430                         if (to_free[j])
 431                                 ++j;
 432                 }
 433
 434                 mutex_unlock(&nf_hook_mutex);
 435
 436                 if (j) {
 437                         unsigned int nfq;
 438
 439                         synchronize_net();
 440
 441                         /* need 2nd synchronize_net() if nfqueue is used, skb
 442                          * can get reinjected right before nf_queue_hook_drop()
 443                          */
 444                         nfq = nf_queue_nf_hook_drop(net);
 445                         if (nfq)
 446                                 synchronize_net();
 447
 448                         for (i = 0; i < j; i++)
 449                                 kvfree(to_free[i]);
 450                 }
 451
 452                 reg += n;
 453                 hookcount -= n;
 454         } while (hookcount > 0);
 455 }
 456 EXPORT_SYMBOL(nf_unregister_net_hooks);
 457
 458 /* Returns 1 if okfn() needs to be executed by the caller,
 459  * -EPERM for NF_DROP, 0 otherwise.  Caller must hold rcu_read_lock. */
 460 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state,
 461                  const struct nf_hook_entries *e, unsigned int s)
 462 {
 463         unsigned int verdict;
 464         int ret;
 465
 466         for (; s < e->num_hook_entries; s++) {
 467                 verdict = nf_hook_entry_hookfn(&e->hooks[s], skb, state);
 468                 switch (verdict & NF_VERDICT_MASK) {
 469                 case NF_ACCEPT:
 470                         break;
 471                 case NF_DROP:
 472                         kfree_skb(skb);
 473                         ret = NF_DROP_GETERR(verdict);
 474                         if (ret == 0)
 475                                 ret = -EPERM;
 476                         return ret;
 477                 case NF_QUEUE:
 478                         ret = nf_queue(skb, state, e, s, verdict);
 479                         if (ret == 1)
 480                                 continue;
 481                         return ret;
 482                 default:
 483                         /* Implicit handling for NF_STOLEN, as well as any other
 484                          * non conventional verdicts.
 485                          */
 486                         return 0;
 487                 }
 488         }
 489
 490         return 1;
 491 }
 492 EXPORT_SYMBOL(nf_hook_slow);
 493
 494
 495 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 496 {
 497         if (writable_len > skb->len)
 498                 return 0;
 499
 500         /* Not exclusive use of packet?  Must copy. */
 501         if (!skb_cloned(skb)) {
 502                 if (writable_len <= skb_headlen(skb))
 503                         return 1;
 504         } else if (skb_clone_writable(skb, writable_len))
 505                 return 1;
 506
 507         if (writable_len <= skb_headlen(skb))
 508                 writable_len = 0;
 509         else
 510                 writable_len -= skb_headlen(skb);
 511
 512         return !!__pskb_pull_tail(skb, writable_len);
 513 }
 514 EXPORT_SYMBOL(skb_make_writable);
 515
 516 /* This needs to be compiled in any case to avoid dependencies between the
 517  * nfnetlink_queue code and nf_conntrack.
 518  */
 519 struct nfnl_ct_hook __rcu *nfnl_ct_hook __read_mostly;
 520 EXPORT_SYMBOL_GPL(nfnl_ct_hook);
 521
 522 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 523 /* This does not belong here, but locally generated errors need it if connection
 524    tracking in use: without this, connection may not be in hash table, and hence
 525    manufactured ICMP or RST packets will not be associated with it. */
 526 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 527                 __rcu __read_mostly;
 528 EXPORT_SYMBOL(ip_ct_attach);
 529
 530 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 531 {
 532         void (*attach)(struct sk_buff *, const struct sk_buff *);
 533
 534         if (skb->_nfct) {
 535                 rcu_read_lock();
 536                 attach = rcu_dereference(ip_ct_attach);
 537                 if (attach)
 538                         attach(new, skb);
 539                 rcu_read_unlock();
 540         }
 541 }
 542 EXPORT_SYMBOL(nf_ct_attach);
 543
 544 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 545 EXPORT_SYMBOL(nf_ct_destroy);
 546
 547 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 548 {
 549         void (*destroy)(struct nf_conntrack *);
 550
 551         rcu_read_lock();
 552         destroy = rcu_dereference(nf_ct_destroy);
 553         BUG_ON(destroy == NULL);
 554         destroy(nfct);
 555         rcu_read_unlock();
 556 }
 557 EXPORT_SYMBOL(nf_conntrack_destroy);
 558
 559 /* Built-in default zone used e.g. by modules. */
 560 const struct nf_conntrack_zone nf_ct_zone_dflt = {
 561         .id     = NF_CT_DEFAULT_ZONE_ID,
 562         .dir    = NF_CT_DEFAULT_ZONE_DIR,
 563 };
 564 EXPORT_SYMBOL_GPL(nf_ct_zone_dflt);
 565 #endif /* CONFIG_NF_CONNTRACK */
 566
 567 #ifdef CONFIG_NF_NAT_NEEDED
 568 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 569 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 570 #endif
 571
 572 static int __net_init netfilter_net_init(struct net *net)
 573 {
 574         int i, h;
 575
 576         for (i = 0; i < ARRAY_SIZE(net->nf.hooks); i++) {
 577                 for (h = 0; h < NF_MAX_HOOKS; h++)
 578                         RCU_INIT_POINTER(net->nf.hooks[i][h], NULL);
 579         }
 580
 581 #ifdef CONFIG_PROC_FS
 582         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 583                                                 net->proc_net);
 584         if (!net->nf.proc_netfilter) {
 585                 if (!net_eq(net, &init_net))
 586                         pr_err("cannot create netfilter proc entry");
 587
 588                 return -ENOMEM;
 589         }
 590 #endif
 591
 592         return 0;
 593 }
 594
 595 static void __net_exit netfilter_net_exit(struct net *net)
 596 {
 597         remove_proc_entry("netfilter", net->proc_net);
 598 }
 599
 600 static struct pernet_operations netfilter_net_ops = {
 601         .init = netfilter_net_init,
 602         .exit = netfilter_net_exit,
 603 };
 604
 605 int __init netfilter_init(void)
 606 {
 607         int ret;
 608
 609         ret = register_pernet_subsys(&netfilter_net_ops);
 610         if (ret < 0)
 611                 goto err;
 612
 613         ret = netfilter_log_init();
 614         if (ret < 0)
 615                 goto err_pernet;
 616
 617         return 0;
 618 err_pernet:
 619         unregister_pernet_subsys(&netfilter_net_ops);
 620 err:
 621         return ret;
 622 }