net/netfilter/nf_conntrack_core.c

   1 /* Connection state tracking for netfilter.  This is separated from,
   2    but required by, the NAT layer; it can also be used by an iptables
   3    extension. */
   4
   5 /* (C) 1999-2001 Paul `Rusty' Russell
   6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License version 2 as
  11  * published by the Free Software Foundation.
  12  */
  13
  14 #include <linux/types.h>
  15 #include <linux/netfilter.h>
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/proc_fs.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/stddef.h>
  21 #include <linux/slab.h>
  22 #include <linux/random.h>
  23 #include <linux/jhash.h>
  24 #include <linux/err.h>
  25 #include <linux/percpu.h>
  26 #include <linux/moduleparam.h>
  27 #include <linux/notifier.h>
  28 #include <linux/kernel.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/socket.h>
  31 #include <linux/mm.h>
  32
  33 #include <net/netfilter/nf_conntrack.h>
  34 #include <net/netfilter/nf_conntrack_l3proto.h>
  35 #include <net/netfilter/nf_conntrack_l4proto.h>
  36 #include <net/netfilter/nf_conntrack_expect.h>
  37 #include <net/netfilter/nf_conntrack_helper.h>
  38 #include <net/netfilter/nf_conntrack_core.h>
  39 #include <net/netfilter/nf_conntrack_extend.h>
  40
  41 #define NF_CONNTRACK_VERSION    "0.5.0"
  42
  43 #if 0
  44 #define DEBUGP printk
  45 #else
  46 #define DEBUGP(format, args...)
  47 #endif
  48
  49 DEFINE_RWLOCK(nf_conntrack_lock);
  50 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  51
  52 /* nf_conntrack_standalone needs this */
  53 atomic_t nf_conntrack_count = ATOMIC_INIT(0);
  54 EXPORT_SYMBOL_GPL(nf_conntrack_count);
  55
  56 unsigned int nf_conntrack_htable_size __read_mostly;
  57 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  58
  59 int nf_conntrack_max __read_mostly;
  60 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  61
  62 struct hlist_head *nf_conntrack_hash __read_mostly;
  63 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
  64
  65 struct nf_conn nf_conntrack_untracked __read_mostly;
  66 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
  67
  68 unsigned int nf_ct_log_invalid __read_mostly;
  69 HLIST_HEAD(unconfirmed);
  70 static int nf_conntrack_vmalloc __read_mostly;
  71 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
  72 static unsigned int nf_conntrack_next_id;
  73
  74 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
  75 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
  76
  77 static int nf_conntrack_hash_rnd_initted;
  78 static unsigned int nf_conntrack_hash_rnd;
  79
  80 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
  81                                   unsigned int size, unsigned int rnd)
  82 {
  83         unsigned int a, b;
  84
  85         a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all),
  86                    (tuple->src.l3num << 16) | tuple->dst.protonum);
  87         b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
  88                    (tuple->src.u.all << 16) | tuple->dst.u.all);
  89
  90         return jhash_2words(a, b, rnd) % size;
  91 }
  92
  93 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
  94 {
  95         return __hash_conntrack(tuple, nf_conntrack_htable_size,
  96                                 nf_conntrack_hash_rnd);
  97 }
  98
  99 int
 100 nf_ct_get_tuple(const struct sk_buff *skb,
 101                 unsigned int nhoff,
 102                 unsigned int dataoff,
 103                 u_int16_t l3num,
 104                 u_int8_t protonum,
 105                 struct nf_conntrack_tuple *tuple,
 106                 const struct nf_conntrack_l3proto *l3proto,
 107                 const struct nf_conntrack_l4proto *l4proto)
 108 {
 109         NF_CT_TUPLE_U_BLANK(tuple);
 110
 111         tuple->src.l3num = l3num;
 112         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 113                 return 0;
 114
 115         tuple->dst.protonum = protonum;
 116         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 117
 118         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 119 }
 120 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 121
 122 int
 123 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 124                    const struct nf_conntrack_tuple *orig,
 125                    const struct nf_conntrack_l3proto *l3proto,
 126                    const struct nf_conntrack_l4proto *l4proto)
 127 {
 128         NF_CT_TUPLE_U_BLANK(inverse);
 129
 130         inverse->src.l3num = orig->src.l3num;
 131         if (l3proto->invert_tuple(inverse, orig) == 0)
 132                 return 0;
 133
 134         inverse->dst.dir = !orig->dst.dir;
 135
 136         inverse->dst.protonum = orig->dst.protonum;
 137         return l4proto->invert_tuple(inverse, orig);
 138 }
 139 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 140
 141 static void
 142 clean_from_lists(struct nf_conn *ct)
 143 {
 144         DEBUGP("clean_from_lists(%p)\n", ct);
 145         hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 146         hlist_del(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
 147
 148         /* Destroy all pending expectations */
 149         nf_ct_remove_expectations(ct);
 150 }
 151
 152 static void
 153 destroy_conntrack(struct nf_conntrack *nfct)
 154 {
 155         struct nf_conn *ct = (struct nf_conn *)nfct;
 156         struct nf_conntrack_l4proto *l4proto;
 157
 158         DEBUGP("destroy_conntrack(%p)\n", ct);
 159         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 160         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 161
 162         nf_conntrack_event(IPCT_DESTROY, ct);
 163         set_bit(IPS_DYING_BIT, &ct->status);
 164
 165         /* To make sure we don't get any weird locking issues here:
 166          * destroy_conntrack() MUST NOT be called with a write lock
 167          * to nf_conntrack_lock!!! -HW */
 168         rcu_read_lock();
 169         l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num,
 170                                        ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
 171         if (l4proto && l4proto->destroy)
 172                 l4proto->destroy(ct);
 173
 174         nf_ct_ext_destroy(ct);
 175
 176         rcu_read_unlock();
 177
 178         write_lock_bh(&nf_conntrack_lock);
 179         /* Expectations will have been removed in clean_from_lists,
 180          * except TFTP can create an expectation on the first packet,
 181          * before connection is in the list, so we need to clean here,
 182          * too. */
 183         nf_ct_remove_expectations(ct);
 184
 185         /* We overload first tuple to link into unconfirmed list. */
 186         if (!nf_ct_is_confirmed(ct)) {
 187                 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
 188                 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 189         }
 190
 191         NF_CT_STAT_INC(delete);
 192         write_unlock_bh(&nf_conntrack_lock);
 193
 194         if (ct->master)
 195                 nf_ct_put(ct->master);
 196
 197         DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct);
 198         nf_conntrack_free(ct);
 199 }
 200
 201 static void death_by_timeout(unsigned long ul_conntrack)
 202 {
 203         struct nf_conn *ct = (void *)ul_conntrack;
 204         struct nf_conn_help *help = nfct_help(ct);
 205         struct nf_conntrack_helper *helper;
 206
 207         if (help) {
 208                 rcu_read_lock();
 209                 helper = rcu_dereference(help->helper);
 210                 if (helper && helper->destroy)
 211                         helper->destroy(ct);
 212                 rcu_read_unlock();
 213         }
 214
 215         write_lock_bh(&nf_conntrack_lock);
 216         /* Inside lock so preempt is disabled on module removal path.
 217          * Otherwise we can get spurious warnings. */
 218         NF_CT_STAT_INC(delete_list);
 219         clean_from_lists(ct);
 220         write_unlock_bh(&nf_conntrack_lock);
 221         nf_ct_put(ct);
 222 }
 223
 224 struct nf_conntrack_tuple_hash *
 225 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple,
 226                     const struct nf_conn *ignored_conntrack)
 227 {
 228         struct nf_conntrack_tuple_hash *h;
 229         struct hlist_node *n;
 230         unsigned int hash = hash_conntrack(tuple);
 231
 232         hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) {
 233                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
 234                     nf_ct_tuple_equal(tuple, &h->tuple)) {
 235                         NF_CT_STAT_INC(found);
 236                         return h;
 237                 }
 238                 NF_CT_STAT_INC(searched);
 239         }
 240
 241         return NULL;
 242 }
 243 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 244
 245 /* Find a connection corresponding to a tuple. */
 246 struct nf_conntrack_tuple_hash *
 247 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple,
 248                       const struct nf_conn *ignored_conntrack)
 249 {
 250         struct nf_conntrack_tuple_hash *h;
 251
 252         read_lock_bh(&nf_conntrack_lock);
 253         h = __nf_conntrack_find(tuple, ignored_conntrack);
 254         if (h)
 255                 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
 256         read_unlock_bh(&nf_conntrack_lock);
 257
 258         return h;
 259 }
 260 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 261
 262 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 263                                        unsigned int hash,
 264                                        unsigned int repl_hash)
 265 {
 266         ct->id = ++nf_conntrack_next_id;
 267         hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
 268                        &nf_conntrack_hash[hash]);
 269         hlist_add_head(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
 270                        &nf_conntrack_hash[repl_hash]);
 271 }
 272
 273 void nf_conntrack_hash_insert(struct nf_conn *ct)
 274 {
 275         unsigned int hash, repl_hash;
 276
 277         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 278         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 279
 280         write_lock_bh(&nf_conntrack_lock);
 281         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 282         write_unlock_bh(&nf_conntrack_lock);
 283 }
 284 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 285
 286 /* Confirm a connection given skb; places it in hash table */
 287 int
 288 __nf_conntrack_confirm(struct sk_buff **pskb)
 289 {
 290         unsigned int hash, repl_hash;
 291         struct nf_conntrack_tuple_hash *h;
 292         struct nf_conn *ct;
 293         struct nf_conn_help *help;
 294         struct hlist_node *n;
 295         enum ip_conntrack_info ctinfo;
 296
 297         ct = nf_ct_get(*pskb, &ctinfo);
 298
 299         /* ipt_REJECT uses nf_conntrack_attach to attach related
 300            ICMP/TCP RST packets in other direction.  Actual packet
 301            which created connection will be IP_CT_NEW or for an
 302            expected connection, IP_CT_RELATED. */
 303         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 304                 return NF_ACCEPT;
 305
 306         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 307         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 308
 309         /* We're not in hash table, and we refuse to set up related
 310            connections for unconfirmed conns.  But packet copies and
 311            REJECT will give spurious warnings here. */
 312         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 313
 314         /* No external references means noone else could have
 315            confirmed us. */
 316         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 317         DEBUGP("Confirming conntrack %p\n", ct);
 318
 319         write_lock_bh(&nf_conntrack_lock);
 320
 321         /* See if there's one in the list already, including reverse:
 322            NAT could have grabbed it without realizing, since we're
 323            not in the hash.  If there is, we lost race. */
 324         hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode)
 325                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 326                                       &h->tuple))
 327                         goto out;
 328         hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode)
 329                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 330                                       &h->tuple))
 331                         goto out;
 332
 333         /* Remove from unconfirmed list */
 334         hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 335
 336         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 337         /* Timer relative to confirmation time, not original
 338            setting time, otherwise we'd get timer wrap in
 339            weird delay cases. */
 340         ct->timeout.expires += jiffies;
 341         add_timer(&ct->timeout);
 342         atomic_inc(&ct->ct_general.use);
 343         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 344         NF_CT_STAT_INC(insert);
 345         write_unlock_bh(&nf_conntrack_lock);
 346         help = nfct_help(ct);
 347         if (help && help->helper)
 348                 nf_conntrack_event_cache(IPCT_HELPER, *pskb);
 349 #ifdef CONFIG_NF_NAT_NEEDED
 350         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
 351             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
 352                 nf_conntrack_event_cache(IPCT_NATINFO, *pskb);
 353 #endif
 354         nf_conntrack_event_cache(master_ct(ct) ?
 355                                  IPCT_RELATED : IPCT_NEW, *pskb);
 356         return NF_ACCEPT;
 357
 358 out:
 359         NF_CT_STAT_INC(insert_failed);
 360         write_unlock_bh(&nf_conntrack_lock);
 361         return NF_DROP;
 362 }
 363 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 364
 365 /* Returns true if a connection correspondings to the tuple (required
 366    for NAT). */
 367 int
 368 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 369                          const struct nf_conn *ignored_conntrack)
 370 {
 371         struct nf_conntrack_tuple_hash *h;
 372
 373         read_lock_bh(&nf_conntrack_lock);
 374         h = __nf_conntrack_find(tuple, ignored_conntrack);
 375         read_unlock_bh(&nf_conntrack_lock);
 376
 377         return h != NULL;
 378 }
 379 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 380
 381 /* There's a small race here where we may free a just-assured
 382    connection.  Too bad: we're in trouble anyway. */
 383 static int early_drop(struct hlist_head *chain)
 384 {
 385         /* Use oldest entry, which is roughly LRU */
 386         struct nf_conntrack_tuple_hash *h;
 387         struct nf_conn *ct = NULL, *tmp;
 388         struct hlist_node *n;
 389         int dropped = 0;
 390
 391         read_lock_bh(&nf_conntrack_lock);
 392         hlist_for_each_entry(h, n, chain, hnode) {
 393                 tmp = nf_ct_tuplehash_to_ctrack(h);
 394                 if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 395                         ct = tmp;
 396         }
 397         if (ct)
 398                 atomic_inc(&ct->ct_general.use);
 399         read_unlock_bh(&nf_conntrack_lock);
 400
 401         if (!ct)
 402                 return dropped;
 403
 404         if (del_timer(&ct->timeout)) {
 405                 death_by_timeout((unsigned long)ct);
 406                 dropped = 1;
 407                 NF_CT_STAT_INC_ATOMIC(early_drop);
 408         }
 409         nf_ct_put(ct);
 410         return dropped;
 411 }
 412
 413 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
 414                                    const struct nf_conntrack_tuple *repl)
 415 {
 416         struct nf_conn *conntrack = NULL;
 417
 418         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 419                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
 420                 nf_conntrack_hash_rnd_initted = 1;
 421         }
 422
 423         /* We don't want any race condition at early drop stage */
 424         atomic_inc(&nf_conntrack_count);
 425
 426         if (nf_conntrack_max
 427             && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
 428                 unsigned int hash = hash_conntrack(orig);
 429                 /* Try dropping from this hash chain. */
 430                 if (!early_drop(&nf_conntrack_hash[hash])) {
 431                         atomic_dec(&nf_conntrack_count);
 432                         if (net_ratelimit())
 433                                 printk(KERN_WARNING
 434                                        "nf_conntrack: table full, dropping"
 435                                        " packet.\n");
 436                         return ERR_PTR(-ENOMEM);
 437                 }
 438         }
 439
 440         conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
 441         if (conntrack == NULL) {
 442                 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack.\n");
 443                 atomic_dec(&nf_conntrack_count);
 444                 return ERR_PTR(-ENOMEM);
 445         }
 446
 447         atomic_set(&conntrack->ct_general.use, 1);
 448         conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 449         conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 450         /* Don't set timer yet: wait for confirmation */
 451         setup_timer(&conntrack->timeout, death_by_timeout,
 452                     (unsigned long)conntrack);
 453
 454         return conntrack;
 455 }
 456 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 457
 458 void nf_conntrack_free(struct nf_conn *conntrack)
 459 {
 460         nf_ct_ext_free(conntrack);
 461         kmem_cache_free(nf_conntrack_cachep, conntrack);
 462         atomic_dec(&nf_conntrack_count);
 463 }
 464 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 465
 466 /* Allocate a new conntrack: we return -ENOMEM if classification
 467    failed due to stress.  Otherwise it really is unclassifiable. */
 468 static struct nf_conntrack_tuple_hash *
 469 init_conntrack(const struct nf_conntrack_tuple *tuple,
 470                struct nf_conntrack_l3proto *l3proto,
 471                struct nf_conntrack_l4proto *l4proto,
 472                struct sk_buff *skb,
 473                unsigned int dataoff)
 474 {
 475         struct nf_conn *conntrack;
 476         struct nf_conn_help *help;
 477         struct nf_conntrack_tuple repl_tuple;
 478         struct nf_conntrack_expect *exp;
 479
 480         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 481                 DEBUGP("Can't invert tuple.\n");
 482                 return NULL;
 483         }
 484
 485         conntrack = nf_conntrack_alloc(tuple, &repl_tuple);
 486         if (conntrack == NULL || IS_ERR(conntrack)) {
 487                 DEBUGP("Can't allocate conntrack.\n");
 488                 return (struct nf_conntrack_tuple_hash *)conntrack;
 489         }
 490
 491         if (!l4proto->new(conntrack, skb, dataoff)) {
 492                 nf_conntrack_free(conntrack);
 493                 DEBUGP("init conntrack: can't track with proto module\n");
 494                 return NULL;
 495         }
 496
 497         write_lock_bh(&nf_conntrack_lock);
 498         exp = find_expectation(tuple);
 499         if (exp) {
 500                 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
 501                         conntrack, exp);
 502                 /* Welcome, Mr. Bond.  We've been expecting you... */
 503                 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
 504                 conntrack->master = exp->master;
 505                 if (exp->helper) {
 506                         help = nf_ct_ext_add(conntrack, NF_CT_EXT_HELPER,
 507                                              GFP_ATOMIC);
 508                         if (help)
 509                                 rcu_assign_pointer(help->helper, exp->helper);
 510                         else
 511                                 DEBUGP("failed to add helper extension area");
 512                 }
 513
 514 #ifdef CONFIG_NF_CONNTRACK_MARK
 515                 conntrack->mark = exp->master->mark;
 516 #endif
 517 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 518                 conntrack->secmark = exp->master->secmark;
 519 #endif
 520                 nf_conntrack_get(&conntrack->master->ct_general);
 521                 NF_CT_STAT_INC(expect_new);
 522         } else {
 523                 struct nf_conntrack_helper *helper;
 524
 525                 helper = __nf_ct_helper_find(&repl_tuple);
 526                 if (helper) {
 527                         help = nf_ct_ext_add(conntrack, NF_CT_EXT_HELPER,
 528                                              GFP_ATOMIC);
 529                         if (help)
 530                                 /* not in hash table yet, so not strictly
 531                                    necessary */
 532                                 rcu_assign_pointer(help->helper, helper);
 533                         else
 534                                 DEBUGP("failed to add helper extension area");
 535                 }
 536                 NF_CT_STAT_INC(new);
 537         }
 538
 539         /* Overload tuple linked list to put us in unconfirmed list. */
 540         hlist_add_head(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
 541                        &unconfirmed);
 542
 543         write_unlock_bh(&nf_conntrack_lock);
 544
 545         if (exp) {
 546                 if (exp->expectfn)
 547                         exp->expectfn(conntrack, exp);
 548                 nf_conntrack_expect_put(exp);
 549         }
 550
 551         return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
 552 }
 553
 554 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 555 static inline struct nf_conn *
 556 resolve_normal_ct(struct sk_buff *skb,
 557                   unsigned int dataoff,
 558                   u_int16_t l3num,
 559                   u_int8_t protonum,
 560                   struct nf_conntrack_l3proto *l3proto,
 561                   struct nf_conntrack_l4proto *l4proto,
 562                   int *set_reply,
 563                   enum ip_conntrack_info *ctinfo)
 564 {
 565         struct nf_conntrack_tuple tuple;
 566         struct nf_conntrack_tuple_hash *h;
 567         struct nf_conn *ct;
 568
 569         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 570                              dataoff, l3num, protonum, &tuple, l3proto,
 571                              l4proto)) {
 572                 DEBUGP("resolve_normal_ct: Can't get tuple\n");
 573                 return NULL;
 574         }
 575
 576         /* look for tuple match */
 577         h = nf_conntrack_find_get(&tuple, NULL);
 578         if (!h) {
 579                 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
 580                 if (!h)
 581                         return NULL;
 582                 if (IS_ERR(h))
 583                         return (void *)h;
 584         }
 585         ct = nf_ct_tuplehash_to_ctrack(h);
 586
 587         /* It exists; we have (non-exclusive) reference. */
 588         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 589                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 590                 /* Please set reply bit if this packet OK */
 591                 *set_reply = 1;
 592         } else {
 593                 /* Once we've had two way comms, always ESTABLISHED. */
 594                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 595                         DEBUGP("nf_conntrack_in: normal packet for %p\n", ct);
 596                         *ctinfo = IP_CT_ESTABLISHED;
 597                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 598                         DEBUGP("nf_conntrack_in: related packet for %p\n", ct);
 599                         *ctinfo = IP_CT_RELATED;
 600                 } else {
 601                         DEBUGP("nf_conntrack_in: new packet for %p\n", ct);
 602                         *ctinfo = IP_CT_NEW;
 603                 }
 604                 *set_reply = 0;
 605         }
 606         skb->nfct = &ct->ct_general;
 607         skb->nfctinfo = *ctinfo;
 608         return ct;
 609 }
 610
 611 unsigned int
 612 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb)
 613 {
 614         struct nf_conn *ct;
 615         enum ip_conntrack_info ctinfo;
 616         struct nf_conntrack_l3proto *l3proto;
 617         struct nf_conntrack_l4proto *l4proto;
 618         unsigned int dataoff;
 619         u_int8_t protonum;
 620         int set_reply = 0;
 621         int ret;
 622
 623         /* Previously seen (loopback or untracked)?  Ignore. */
 624         if ((*pskb)->nfct) {
 625                 NF_CT_STAT_INC_ATOMIC(ignore);
 626                 return NF_ACCEPT;
 627         }
 628
 629         /* rcu_read_lock()ed by nf_hook_slow */
 630         l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
 631
 632         if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) {
 633                 DEBUGP("not prepared to track yet or error occured\n");
 634                 return -ret;
 635         }
 636
 637         l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
 638
 639         /* It may be an special packet, error, unclean...
 640          * inverse of the return code tells to the netfilter
 641          * core what to do with the packet. */
 642         if (l4proto->error != NULL &&
 643             (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
 644                 NF_CT_STAT_INC_ATOMIC(error);
 645                 NF_CT_STAT_INC_ATOMIC(invalid);
 646                 return -ret;
 647         }
 648
 649         ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto,
 650                                &set_reply, &ctinfo);
 651         if (!ct) {
 652                 /* Not valid part of a connection */
 653                 NF_CT_STAT_INC_ATOMIC(invalid);
 654                 return NF_ACCEPT;
 655         }
 656
 657         if (IS_ERR(ct)) {
 658                 /* Too stressed to deal. */
 659                 NF_CT_STAT_INC_ATOMIC(drop);
 660                 return NF_DROP;
 661         }
 662
 663         NF_CT_ASSERT((*pskb)->nfct);
 664
 665         ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum);
 666         if (ret < 0) {
 667                 /* Invalid: inverse of the return code tells
 668                  * the netfilter core what to do */
 669                 DEBUGP("nf_conntrack_in: Can't track with proto module\n");
 670                 nf_conntrack_put((*pskb)->nfct);
 671                 (*pskb)->nfct = NULL;
 672                 NF_CT_STAT_INC_ATOMIC(invalid);
 673                 return -ret;
 674         }
 675
 676         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 677                 nf_conntrack_event_cache(IPCT_STATUS, *pskb);
 678
 679         return ret;
 680 }
 681 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 682
 683 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 684                          const struct nf_conntrack_tuple *orig)
 685 {
 686         int ret;
 687
 688         rcu_read_lock();
 689         ret = nf_ct_invert_tuple(inverse, orig,
 690                                  __nf_ct_l3proto_find(orig->src.l3num),
 691                                  __nf_ct_l4proto_find(orig->src.l3num,
 692                                                       orig->dst.protonum));
 693         rcu_read_unlock();
 694         return ret;
 695 }
 696 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 697
 698 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 699    implicitly racy: see __nf_conntrack_confirm */
 700 void nf_conntrack_alter_reply(struct nf_conn *ct,
 701                               const struct nf_conntrack_tuple *newreply)
 702 {
 703         struct nf_conn_help *help = nfct_help(ct);
 704         struct nf_conntrack_helper *helper;
 705
 706         write_lock_bh(&nf_conntrack_lock);
 707         /* Should be unconfirmed, so not in hash table yet */
 708         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 709
 710         DEBUGP("Altering reply tuple of %p to ", ct);
 711         NF_CT_DUMP_TUPLE(newreply);
 712
 713         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 714         if (ct->master || (help && help->expecting != 0))
 715                 goto out;
 716
 717         helper = __nf_ct_helper_find(newreply);
 718         if (helper == NULL) {
 719                 if (help)
 720                         rcu_assign_pointer(help->helper, NULL);
 721                 goto out;
 722         }
 723
 724         if (help == NULL) {
 725                 help = nf_ct_ext_add(ct, NF_CT_EXT_HELPER, GFP_ATOMIC);
 726                 if (help == NULL) {
 727                         DEBUGP("failed to add helper extension area");
 728                         goto out;
 729                 }
 730         } else {
 731                 memset(&help->help, 0, sizeof(help->help));
 732         }
 733
 734         rcu_assign_pointer(help->helper, helper);
 735 out:
 736         write_unlock_bh(&nf_conntrack_lock);
 737 }
 738 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 739
 740 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 741 void __nf_ct_refresh_acct(struct nf_conn *ct,
 742                           enum ip_conntrack_info ctinfo,
 743                           const struct sk_buff *skb,
 744                           unsigned long extra_jiffies,
 745                           int do_acct)
 746 {
 747         int event = 0;
 748
 749         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 750         NF_CT_ASSERT(skb);
 751
 752         write_lock_bh(&nf_conntrack_lock);
 753
 754         /* Only update if this is not a fixed timeout */
 755         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) {
 756                 write_unlock_bh(&nf_conntrack_lock);
 757                 return;
 758         }
 759
 760         /* If not in hash table, timer will not be active yet */
 761         if (!nf_ct_is_confirmed(ct)) {
 762                 ct->timeout.expires = extra_jiffies;
 763                 event = IPCT_REFRESH;
 764         } else {
 765                 unsigned long newtime = jiffies + extra_jiffies;
 766
 767                 /* Only update the timeout if the new timeout is at least
 768                    HZ jiffies from the old timeout. Need del_timer for race
 769                    avoidance (may already be dying). */
 770                 if (newtime - ct->timeout.expires >= HZ
 771                     && del_timer(&ct->timeout)) {
 772                         ct->timeout.expires = newtime;
 773                         add_timer(&ct->timeout);
 774                         event = IPCT_REFRESH;
 775                 }
 776         }
 777
 778 #ifdef CONFIG_NF_CT_ACCT
 779         if (do_acct) {
 780                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
 781                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
 782                         skb->len - skb_network_offset(skb);
 783
 784                 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
 785                     || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
 786                         event |= IPCT_COUNTER_FILLING;
 787         }
 788 #endif
 789
 790         write_unlock_bh(&nf_conntrack_lock);
 791
 792         /* must be unlocked when calling event cache */
 793         if (event)
 794                 nf_conntrack_event_cache(event, skb);
 795 }
 796 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 797
 798 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 799
 800 #include <linux/netfilter/nfnetlink.h>
 801 #include <linux/netfilter/nfnetlink_conntrack.h>
 802 #include <linux/mutex.h>
 803
 804
 805 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 806  * in ip_conntrack_core, since we don't want the protocols to autoload
 807  * or depend on ctnetlink */
 808 int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb,
 809                                const struct nf_conntrack_tuple *tuple)
 810 {
 811         NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t),
 812                 &tuple->src.u.tcp.port);
 813         NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t),
 814                 &tuple->dst.u.tcp.port);
 815         return 0;
 816
 817 nfattr_failure:
 818         return -1;
 819 }
 820 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr);
 821
 822 static const size_t cta_min_proto[CTA_PROTO_MAX] = {
 823         [CTA_PROTO_SRC_PORT-1]  = sizeof(u_int16_t),
 824         [CTA_PROTO_DST_PORT-1]  = sizeof(u_int16_t)
 825 };
 826
 827 int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[],
 828                                struct nf_conntrack_tuple *t)
 829 {
 830         if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1])
 831                 return -EINVAL;
 832
 833         if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
 834                 return -EINVAL;
 835
 836         t->src.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]);
 837         t->dst.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]);
 838
 839         return 0;
 840 }
 841 EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple);
 842 #endif
 843
 844 /* Used by ipt_REJECT and ip6t_REJECT. */
 845 void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 846 {
 847         struct nf_conn *ct;
 848         enum ip_conntrack_info ctinfo;
 849
 850         /* This ICMP is in reverse direction to the packet which caused it */
 851         ct = nf_ct_get(skb, &ctinfo);
 852         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 853                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 854         else
 855                 ctinfo = IP_CT_RELATED;
 856
 857         /* Attach to new skbuff, and increment count */
 858         nskb->nfct = &ct->ct_general;
 859         nskb->nfctinfo = ctinfo;
 860         nf_conntrack_get(nskb->nfct);
 861 }
 862 EXPORT_SYMBOL_GPL(__nf_conntrack_attach);
 863
 864 static inline int
 865 do_iter(const struct nf_conntrack_tuple_hash *i,
 866         int (*iter)(struct nf_conn *i, void *data),
 867         void *data)
 868 {
 869         return iter(nf_ct_tuplehash_to_ctrack(i), data);
 870 }
 871
 872 /* Bring out ya dead! */
 873 static struct nf_conn *
 874 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
 875                 void *data, unsigned int *bucket)
 876 {
 877         struct nf_conntrack_tuple_hash *h;
 878         struct nf_conn *ct;
 879         struct hlist_node *n;
 880
 881         write_lock_bh(&nf_conntrack_lock);
 882         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 883                 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
 884                         ct = nf_ct_tuplehash_to_ctrack(h);
 885                         if (iter(ct, data))
 886                                 goto found;
 887                 }
 888         }
 889         hlist_for_each_entry(h, n, &unconfirmed, hnode) {
 890                 ct = nf_ct_tuplehash_to_ctrack(h);
 891                 if (iter(ct, data))
 892                         set_bit(IPS_DYING_BIT, &ct->status);
 893         }
 894         write_unlock_bh(&nf_conntrack_lock);
 895         return NULL;
 896 found:
 897         atomic_inc(&ct->ct_general.use);
 898         write_unlock_bh(&nf_conntrack_lock);
 899         return ct;
 900 }
 901
 902 void
 903 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
 904 {
 905         struct nf_conn *ct;
 906         unsigned int bucket = 0;
 907
 908         while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
 909                 /* Time to push up daises... */
 910                 if (del_timer(&ct->timeout))
 911                         death_by_timeout((unsigned long)ct);
 912                 /* ... else the timer will get him soon. */
 913
 914                 nf_ct_put(ct);
 915         }
 916 }
 917 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
 918
 919 static int kill_all(struct nf_conn *i, void *data)
 920 {
 921         return 1;
 922 }
 923
 924 static void free_conntrack_hash(struct hlist_head *hash, int vmalloced,
 925                                 int size)
 926 {
 927         if (vmalloced)
 928                 vfree(hash);
 929         else
 930                 free_pages((unsigned long)hash,
 931                            get_order(sizeof(struct hlist_head) * size));
 932 }
 933
 934 void nf_conntrack_flush(void)
 935 {
 936         nf_ct_iterate_cleanup(kill_all, NULL);
 937 }
 938 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
 939
 940 /* Mishearing the voices in his head, our hero wonders how he's
 941    supposed to kill the mall. */
 942 void nf_conntrack_cleanup(void)
 943 {
 944         rcu_assign_pointer(ip_ct_attach, NULL);
 945
 946         /* This makes sure all current packets have passed through
 947            netfilter framework.  Roll on, two-stage module
 948            delete... */
 949         synchronize_net();
 950
 951         nf_ct_event_cache_flush();
 952  i_see_dead_people:
 953         nf_conntrack_flush();
 954         if (atomic_read(&nf_conntrack_count) != 0) {
 955                 schedule();
 956                 goto i_see_dead_people;
 957         }
 958         /* wait until all references to nf_conntrack_untracked are dropped */
 959         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
 960                 schedule();
 961
 962         rcu_assign_pointer(nf_ct_destroy, NULL);
 963
 964         kmem_cache_destroy(nf_conntrack_cachep);
 965         kmem_cache_destroy(nf_conntrack_expect_cachep);
 966         free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
 967                             nf_conntrack_htable_size);
 968
 969         nf_conntrack_proto_fini();
 970         nf_conntrack_helper_fini();
 971 }
 972
 973 static struct hlist_head *alloc_hashtable(int *sizep, int *vmalloced)
 974 {
 975         struct hlist_head *hash;
 976         unsigned int size, i;
 977
 978         *vmalloced = 0;
 979
 980         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
 981         hash = (void*)__get_free_pages(GFP_KERNEL,
 982                                        get_order(sizeof(struct hlist_head)
 983                                                  * size));
 984         if (!hash) {
 985                 *vmalloced = 1;
 986                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
 987                 hash = vmalloc(sizeof(struct hlist_head) * size);
 988         }
 989
 990         if (hash)
 991                 for (i = 0; i < size; i++)
 992                         INIT_HLIST_HEAD(&hash[i]);
 993
 994         return hash;
 995 }
 996
 997 int set_hashsize(const char *val, struct kernel_param *kp)
 998 {
 999         int i, bucket, hashsize, vmalloced;
1000         int old_vmalloced, old_size;
1001         int rnd;
1002         struct hlist_head *hash, *old_hash;
1003         struct nf_conntrack_tuple_hash *h;
1004
1005         /* On boot, we can set this without any fancy locking. */
1006         if (!nf_conntrack_htable_size)
1007                 return param_set_uint(val, kp);
1008
1009         hashsize = simple_strtol(val, NULL, 0);
1010         if (!hashsize)
1011                 return -EINVAL;
1012
1013         hash = alloc_hashtable(&hashsize, &vmalloced);
1014         if (!hash)
1015                 return -ENOMEM;
1016
1017         /* We have to rehahs for the new table anyway, so we also can
1018          * use a newrandom seed */
1019         get_random_bytes(&rnd, 4);
1020
1021         write_lock_bh(&nf_conntrack_lock);
1022         for (i = 0; i < nf_conntrack_htable_size; i++) {
1023                 while (!hlist_empty(&nf_conntrack_hash[i])) {
1024                         h = hlist_entry(nf_conntrack_hash[i].first,
1025                                         struct nf_conntrack_tuple_hash, hnode);
1026                         hlist_del(&h->hnode);
1027                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1028                         hlist_add_head(&h->hnode, &hash[bucket]);
1029                 }
1030         }
1031         old_size = nf_conntrack_htable_size;
1032         old_vmalloced = nf_conntrack_vmalloc;
1033         old_hash = nf_conntrack_hash;
1034
1035         nf_conntrack_htable_size = hashsize;
1036         nf_conntrack_vmalloc = vmalloced;
1037         nf_conntrack_hash = hash;
1038         nf_conntrack_hash_rnd = rnd;
1039         write_unlock_bh(&nf_conntrack_lock);
1040
1041         free_conntrack_hash(old_hash, old_vmalloced, old_size);
1042         return 0;
1043 }
1044
1045 module_param_call(hashsize, set_hashsize, param_get_uint,
1046                   &nf_conntrack_htable_size, 0600);
1047
1048 int __init nf_conntrack_init(void)
1049 {
1050         int max_factor = 8;
1051         int ret;
1052
1053         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1054          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1055         if (!nf_conntrack_htable_size) {
1056                 nf_conntrack_htable_size
1057                         = (((num_physpages << PAGE_SHIFT) / 16384)
1058                            / sizeof(struct hlist_head));
1059                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1060                         nf_conntrack_htable_size = 16384;
1061                 if (nf_conntrack_htable_size < 32)
1062                         nf_conntrack_htable_size = 32;
1063
1064                 /* Use a max. factor of four by default to get the same max as
1065                  * with the old struct list_heads. When a table size is given
1066                  * we use the old value of 8 to avoid reducing the max.
1067                  * entries. */
1068                 max_factor = 4;
1069         }
1070         nf_conntrack_hash = alloc_hashtable(&nf_conntrack_htable_size,
1071                                             &nf_conntrack_vmalloc);
1072         if (!nf_conntrack_hash) {
1073                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1074                 goto err_out;
1075         }
1076
1077         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1078
1079         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1080                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1081                nf_conntrack_max);
1082
1083         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1084                                                 sizeof(struct nf_conn),
1085                                                 0, 0, NULL, NULL);
1086         if (!nf_conntrack_cachep) {
1087                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1088                 goto err_free_hash;
1089         }
1090
1091         nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect",
1092                                         sizeof(struct nf_conntrack_expect),
1093                                         0, 0, NULL, NULL);
1094         if (!nf_conntrack_expect_cachep) {
1095                 printk(KERN_ERR "Unable to create nf_expect slab cache\n");
1096                 goto err_free_conntrack_slab;
1097         }
1098
1099         ret = nf_conntrack_proto_init();
1100         if (ret < 0)
1101                 goto out_free_expect_slab;
1102
1103         ret = nf_conntrack_helper_init();
1104         if (ret < 0)
1105                 goto out_fini_proto;
1106
1107         /* For use by REJECT target */
1108         rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach);
1109         rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1110
1111         /* Set up fake conntrack:
1112             - to never be deleted, not in any hashes */
1113         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1114         /*  - and look it like as a confirmed connection */
1115         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1116
1117         return ret;
1118
1119 out_fini_proto:
1120         nf_conntrack_proto_fini();
1121 out_free_expect_slab:
1122         kmem_cache_destroy(nf_conntrack_expect_cachep);
1123 err_free_conntrack_slab:
1124         kmem_cache_destroy(nf_conntrack_cachep);
1125 err_free_hash:
1126         free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
1127                             nf_conntrack_htable_size);
1128 err_out:
1129         return -ENOMEM;
1130 }