SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 void (*nf_conntrack_destroyed)(struct nf_conn *conntrack);
  57 EXPORT_SYMBOL_GPL(nf_conntrack_destroyed);
  58
  59 unsigned int nf_conntrack_htable_size __read_mostly;
  60 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  61
  62 int nf_conntrack_max __read_mostly;
  63 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  64
  65 struct list_head *nf_conntrack_hash __read_mostly;
  66 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
  67
  68 struct nf_conn nf_conntrack_untracked __read_mostly;
  69 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
  70
  71 unsigned int nf_ct_log_invalid __read_mostly;
  72 LIST_HEAD(unconfirmed);
  73 static int nf_conntrack_vmalloc __read_mostly;
  74 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
  75 static unsigned int nf_conntrack_next_id;
  76
  77 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
  78 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
  79
  80 static int nf_conntrack_hash_rnd_initted;
  81 static unsigned int nf_conntrack_hash_rnd;
  82
  83 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
  84                                   unsigned int size, unsigned int rnd)
  85 {
  86         unsigned int a, b;
  87
  88         a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all),
  89                    (tuple->src.l3num << 16) | tuple->dst.protonum);
  90         b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
  91                    (tuple->src.u.all << 16) | tuple->dst.u.all);
  92
  93         return jhash_2words(a, b, rnd) % size;
  94 }
  95
  96 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
  97 {
  98         return __hash_conntrack(tuple, nf_conntrack_htable_size,
  99                                 nf_conntrack_hash_rnd);
 100 }
 101
 102 int
 103 nf_ct_get_tuple(const struct sk_buff *skb,
 104                 unsigned int nhoff,
 105                 unsigned int dataoff,
 106                 u_int16_t l3num,
 107                 u_int8_t protonum,
 108                 struct nf_conntrack_tuple *tuple,
 109                 const struct nf_conntrack_l3proto *l3proto,
 110                 const struct nf_conntrack_l4proto *l4proto)
 111 {
 112         NF_CT_TUPLE_U_BLANK(tuple);
 113
 114         tuple->src.l3num = l3num;
 115         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 116                 return 0;
 117
 118         tuple->dst.protonum = protonum;
 119         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 120
 121         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 122 }
 123 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 124
 125 int
 126 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 127                    const struct nf_conntrack_tuple *orig,
 128                    const struct nf_conntrack_l3proto *l3proto,
 129                    const struct nf_conntrack_l4proto *l4proto)
 130 {
 131         NF_CT_TUPLE_U_BLANK(inverse);
 132
 133         inverse->src.l3num = orig->src.l3num;
 134         if (l3proto->invert_tuple(inverse, orig) == 0)
 135                 return 0;
 136
 137         inverse->dst.dir = !orig->dst.dir;
 138
 139         inverse->dst.protonum = orig->dst.protonum;
 140         return l4proto->invert_tuple(inverse, orig);
 141 }
 142 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 143
 144 static void
 145 clean_from_lists(struct nf_conn *ct)
 146 {
 147         DEBUGP("clean_from_lists(%p)\n", ct);
 148         list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
 149         list_del(&ct->tuplehash[IP_CT_DIR_REPLY].list);
 150
 151         /* Destroy all pending expectations */
 152         nf_ct_remove_expectations(ct);
 153 }
 154
 155 static void
 156 destroy_conntrack(struct nf_conntrack *nfct)
 157 {
 158         struct nf_conn *ct = (struct nf_conn *)nfct;
 159         struct nf_conntrack_l4proto *l4proto;
 160         typeof(nf_conntrack_destroyed) destroyed;
 161
 162         DEBUGP("destroy_conntrack(%p)\n", ct);
 163         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 164         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 165
 166         nf_conntrack_event(IPCT_DESTROY, ct);
 167         set_bit(IPS_DYING_BIT, &ct->status);
 168
 169         /* To make sure we don't get any weird locking issues here:
 170          * destroy_conntrack() MUST NOT be called with a write lock
 171          * to nf_conntrack_lock!!! -HW */
 172         rcu_read_lock();
 173         l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num,
 174                                        ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
 175         if (l4proto && l4proto->destroy)
 176                 l4proto->destroy(ct);
 177
 178         nf_ct_ext_destroy(ct);
 179
 180         destroyed = rcu_dereference(nf_conntrack_destroyed);
 181         if (destroyed)
 182                 destroyed(ct);
 183
 184         rcu_read_unlock();
 185
 186         write_lock_bh(&nf_conntrack_lock);
 187         /* Expectations will have been removed in clean_from_lists,
 188          * except TFTP can create an expectation on the first packet,
 189          * before connection is in the list, so we need to clean here,
 190          * too. */
 191         nf_ct_remove_expectations(ct);
 192
 193         /* We overload first tuple to link into unconfirmed list. */
 194         if (!nf_ct_is_confirmed(ct)) {
 195                 BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list));
 196                 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
 197         }
 198
 199         NF_CT_STAT_INC(delete);
 200         write_unlock_bh(&nf_conntrack_lock);
 201
 202         if (ct->master)
 203                 nf_ct_put(ct->master);
 204
 205         DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct);
 206         nf_conntrack_free(ct);
 207 }
 208
 209 static void death_by_timeout(unsigned long ul_conntrack)
 210 {
 211         struct nf_conn *ct = (void *)ul_conntrack;
 212         struct nf_conn_help *help = nfct_help(ct);
 213         struct nf_conntrack_helper *helper;
 214
 215         if (help) {
 216                 rcu_read_lock();
 217                 helper = rcu_dereference(help->helper);
 218                 if (helper && helper->destroy)
 219                         helper->destroy(ct);
 220                 rcu_read_unlock();
 221         }
 222
 223         write_lock_bh(&nf_conntrack_lock);
 224         /* Inside lock so preempt is disabled on module removal path.
 225          * Otherwise we can get spurious warnings. */
 226         NF_CT_STAT_INC(delete_list);
 227         clean_from_lists(ct);
 228         write_unlock_bh(&nf_conntrack_lock);
 229         nf_ct_put(ct);
 230 }
 231
 232 struct nf_conntrack_tuple_hash *
 233 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple,
 234                     const struct nf_conn *ignored_conntrack)
 235 {
 236         struct nf_conntrack_tuple_hash *h;
 237         unsigned int hash = hash_conntrack(tuple);
 238
 239         list_for_each_entry(h, &nf_conntrack_hash[hash], list) {
 240                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
 241                     nf_ct_tuple_equal(tuple, &h->tuple)) {
 242                         NF_CT_STAT_INC(found);
 243                         return h;
 244                 }
 245                 NF_CT_STAT_INC(searched);
 246         }
 247
 248         return NULL;
 249 }
 250 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 251
 252 /* Find a connection corresponding to a tuple. */
 253 struct nf_conntrack_tuple_hash *
 254 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple,
 255                       const struct nf_conn *ignored_conntrack)
 256 {
 257         struct nf_conntrack_tuple_hash *h;
 258
 259         read_lock_bh(&nf_conntrack_lock);
 260         h = __nf_conntrack_find(tuple, ignored_conntrack);
 261         if (h)
 262                 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
 263         read_unlock_bh(&nf_conntrack_lock);
 264
 265         return h;
 266 }
 267 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 268
 269 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 270                                        unsigned int hash,
 271                                        unsigned int repl_hash)
 272 {
 273         ct->id = ++nf_conntrack_next_id;
 274         list_add(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list,
 275                  &nf_conntrack_hash[hash]);
 276         list_add(&ct->tuplehash[IP_CT_DIR_REPLY].list,
 277                  &nf_conntrack_hash[repl_hash]);
 278 }
 279
 280 void nf_conntrack_hash_insert(struct nf_conn *ct)
 281 {
 282         unsigned int hash, repl_hash;
 283
 284         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 285         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 286
 287         write_lock_bh(&nf_conntrack_lock);
 288         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 289         write_unlock_bh(&nf_conntrack_lock);
 290 }
 291 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 292
 293 /* Confirm a connection given skb; places it in hash table */
 294 int
 295 __nf_conntrack_confirm(struct sk_buff **pskb)
 296 {
 297         unsigned int hash, repl_hash;
 298         struct nf_conntrack_tuple_hash *h;
 299         struct nf_conn *ct;
 300         struct nf_conn_help *help;
 301         enum ip_conntrack_info ctinfo;
 302
 303         ct = nf_ct_get(*pskb, &ctinfo);
 304
 305         /* ipt_REJECT uses nf_conntrack_attach to attach related
 306            ICMP/TCP RST packets in other direction.  Actual packet
 307            which created connection will be IP_CT_NEW or for an
 308            expected connection, IP_CT_RELATED. */
 309         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 310                 return NF_ACCEPT;
 311
 312         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 313         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 314
 315         /* We're not in hash table, and we refuse to set up related
 316            connections for unconfirmed conns.  But packet copies and
 317            REJECT will give spurious warnings here. */
 318         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 319
 320         /* No external references means noone else could have
 321            confirmed us. */
 322         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 323         DEBUGP("Confirming conntrack %p\n", ct);
 324
 325         write_lock_bh(&nf_conntrack_lock);
 326
 327         /* See if there's one in the list already, including reverse:
 328            NAT could have grabbed it without realizing, since we're
 329            not in the hash.  If there is, we lost race. */
 330         list_for_each_entry(h, &nf_conntrack_hash[hash], list)
 331                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 332                                       &h->tuple))
 333                         goto out;
 334         list_for_each_entry(h, &nf_conntrack_hash[repl_hash], list)
 335                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 336                                       &h->tuple))
 337                         goto out;
 338
 339         /* Remove from unconfirmed list */
 340         list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
 341
 342         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 343         /* Timer relative to confirmation time, not original
 344            setting time, otherwise we'd get timer wrap in
 345            weird delay cases. */
 346         ct->timeout.expires += jiffies;
 347         add_timer(&ct->timeout);
 348         atomic_inc(&ct->ct_general.use);
 349         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 350         NF_CT_STAT_INC(insert);
 351         write_unlock_bh(&nf_conntrack_lock);
 352         help = nfct_help(ct);
 353         if (help && help->helper)
 354                 nf_conntrack_event_cache(IPCT_HELPER, *pskb);
 355 #ifdef CONFIG_NF_NAT_NEEDED
 356         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
 357             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
 358                 nf_conntrack_event_cache(IPCT_NATINFO, *pskb);
 359 #endif
 360         nf_conntrack_event_cache(master_ct(ct) ?
 361                                  IPCT_RELATED : IPCT_NEW, *pskb);
 362         return NF_ACCEPT;
 363
 364 out:
 365         NF_CT_STAT_INC(insert_failed);
 366         write_unlock_bh(&nf_conntrack_lock);
 367         return NF_DROP;
 368 }
 369 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 370
 371 /* Returns true if a connection correspondings to the tuple (required
 372    for NAT). */
 373 int
 374 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 375                          const struct nf_conn *ignored_conntrack)
 376 {
 377         struct nf_conntrack_tuple_hash *h;
 378
 379         read_lock_bh(&nf_conntrack_lock);
 380         h = __nf_conntrack_find(tuple, ignored_conntrack);
 381         read_unlock_bh(&nf_conntrack_lock);
 382
 383         return h != NULL;
 384 }
 385 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 386
 387 /* There's a small race here where we may free a just-assured
 388    connection.  Too bad: we're in trouble anyway. */
 389 static int early_drop(struct list_head *chain)
 390 {
 391         /* Traverse backwards: gives us oldest, which is roughly LRU */
 392         struct nf_conntrack_tuple_hash *h;
 393         struct nf_conn *ct = NULL, *tmp;
 394         int dropped = 0;
 395
 396         read_lock_bh(&nf_conntrack_lock);
 397         list_for_each_entry_reverse(h, chain, list) {
 398                 tmp = nf_ct_tuplehash_to_ctrack(h);
 399                 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) {
 400                         ct = tmp;
 401                         atomic_inc(&ct->ct_general.use);
 402                         break;
 403                 }
 404         }
 405         read_unlock_bh(&nf_conntrack_lock);
 406
 407         if (!ct)
 408                 return dropped;
 409
 410         if (del_timer(&ct->timeout)) {
 411                 death_by_timeout((unsigned long)ct);
 412                 dropped = 1;
 413                 NF_CT_STAT_INC_ATOMIC(early_drop);
 414         }
 415         nf_ct_put(ct);
 416         return dropped;
 417 }
 418
 419 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
 420                                    const struct nf_conntrack_tuple *repl)
 421 {
 422         struct nf_conn *conntrack = NULL;
 423
 424         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 425                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
 426                 nf_conntrack_hash_rnd_initted = 1;
 427         }
 428
 429         /* We don't want any race condition at early drop stage */
 430         atomic_inc(&nf_conntrack_count);
 431
 432         if (nf_conntrack_max
 433             && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
 434                 unsigned int hash = hash_conntrack(orig);
 435                 /* Try dropping from this hash chain. */
 436                 if (!early_drop(&nf_conntrack_hash[hash])) {
 437                         atomic_dec(&nf_conntrack_count);
 438                         if (net_ratelimit())
 439                                 printk(KERN_WARNING
 440                                        "nf_conntrack: table full, dropping"
 441                                        " packet.\n");
 442                         return ERR_PTR(-ENOMEM);
 443                 }
 444         }
 445
 446         conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
 447         if (conntrack == NULL) {
 448                 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack.\n");
 449                 atomic_dec(&nf_conntrack_count);
 450                 return ERR_PTR(-ENOMEM);
 451         }
 452
 453         atomic_set(&conntrack->ct_general.use, 1);
 454         conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 455         conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 456         /* Don't set timer yet: wait for confirmation */
 457         setup_timer(&conntrack->timeout, death_by_timeout,
 458                     (unsigned long)conntrack);
 459
 460         return conntrack;
 461 }
 462 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 463
 464 void nf_conntrack_free(struct nf_conn *conntrack)
 465 {
 466         nf_ct_ext_free(conntrack);
 467         kmem_cache_free(nf_conntrack_cachep, conntrack);
 468         atomic_dec(&nf_conntrack_count);
 469 }
 470 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 471
 472 /* Allocate a new conntrack: we return -ENOMEM if classification
 473    failed due to stress.  Otherwise it really is unclassifiable. */
 474 static struct nf_conntrack_tuple_hash *
 475 init_conntrack(const struct nf_conntrack_tuple *tuple,
 476                struct nf_conntrack_l3proto *l3proto,
 477                struct nf_conntrack_l4proto *l4proto,
 478                struct sk_buff *skb,
 479                unsigned int dataoff)
 480 {
 481         struct nf_conn *conntrack;
 482         struct nf_conn_help *help;
 483         struct nf_conntrack_tuple repl_tuple;
 484         struct nf_conntrack_expect *exp;
 485
 486         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 487                 DEBUGP("Can't invert tuple.\n");
 488                 return NULL;
 489         }
 490
 491         conntrack = nf_conntrack_alloc(tuple, &repl_tuple);
 492         if (conntrack == NULL || IS_ERR(conntrack)) {
 493                 DEBUGP("Can't allocate conntrack.\n");
 494                 return (struct nf_conntrack_tuple_hash *)conntrack;
 495         }
 496
 497         if (!l4proto->new(conntrack, skb, dataoff)) {
 498                 nf_conntrack_free(conntrack);
 499                 DEBUGP("init conntrack: can't track with proto module\n");
 500                 return NULL;
 501         }
 502
 503         write_lock_bh(&nf_conntrack_lock);
 504         exp = find_expectation(tuple);
 505         if (exp) {
 506                 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
 507                         conntrack, exp);
 508                 /* Welcome, Mr. Bond.  We've been expecting you... */
 509                 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
 510                 conntrack->master = exp->master;
 511                 if (exp->helper) {
 512                         help = nf_ct_ext_add(conntrack, NF_CT_EXT_HELPER,
 513                                              GFP_ATOMIC);
 514                         if (help)
 515                                 rcu_assign_pointer(help->helper, exp->helper);
 516                         else
 517                                 DEBUGP("failed to add helper extension area");
 518                 }
 519
 520 #ifdef CONFIG_NF_CONNTRACK_MARK
 521                 conntrack->mark = exp->master->mark;
 522 #endif
 523 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 524                 conntrack->secmark = exp->master->secmark;
 525 #endif
 526                 nf_conntrack_get(&conntrack->master->ct_general);
 527                 NF_CT_STAT_INC(expect_new);
 528         } else {
 529                 struct nf_conntrack_helper *helper;
 530
 531                 helper = __nf_ct_helper_find(&repl_tuple);
 532                 if (helper) {
 533                         help = nf_ct_ext_add(conntrack, NF_CT_EXT_HELPER,
 534                                              GFP_ATOMIC);
 535                         if (help)
 536                                 /* not in hash table yet, so not strictly
 537                                    necessary */
 538                                 rcu_assign_pointer(help->helper, helper);
 539                         else
 540                                 DEBUGP("failed to add helper extension area");
 541                 }
 542                 NF_CT_STAT_INC(new);
 543         }
 544
 545         /* Overload tuple linked list to put us in unconfirmed list. */
 546         list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed);
 547
 548         write_unlock_bh(&nf_conntrack_lock);
 549
 550         if (exp) {
 551                 if (exp->expectfn)
 552                         exp->expectfn(conntrack, exp);
 553                 nf_conntrack_expect_put(exp);
 554         }
 555
 556         return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
 557 }
 558
 559 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 560 static inline struct nf_conn *
 561 resolve_normal_ct(struct sk_buff *skb,
 562                   unsigned int dataoff,
 563                   u_int16_t l3num,
 564                   u_int8_t protonum,
 565                   struct nf_conntrack_l3proto *l3proto,
 566                   struct nf_conntrack_l4proto *l4proto,
 567                   int *set_reply,
 568                   enum ip_conntrack_info *ctinfo)
 569 {
 570         struct nf_conntrack_tuple tuple;
 571         struct nf_conntrack_tuple_hash *h;
 572         struct nf_conn *ct;
 573
 574         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 575                              dataoff, l3num, protonum, &tuple, l3proto,
 576                              l4proto)) {
 577                 DEBUGP("resolve_normal_ct: Can't get tuple\n");
 578                 return NULL;
 579         }
 580
 581         /* look for tuple match */
 582         h = nf_conntrack_find_get(&tuple, NULL);
 583         if (!h) {
 584                 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
 585                 if (!h)
 586                         return NULL;
 587                 if (IS_ERR(h))
 588                         return (void *)h;
 589         }
 590         ct = nf_ct_tuplehash_to_ctrack(h);
 591
 592         /* It exists; we have (non-exclusive) reference. */
 593         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 594                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 595                 /* Please set reply bit if this packet OK */
 596                 *set_reply = 1;
 597         } else {
 598                 /* Once we've had two way comms, always ESTABLISHED. */
 599                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 600                         DEBUGP("nf_conntrack_in: normal packet for %p\n", ct);
 601                         *ctinfo = IP_CT_ESTABLISHED;
 602                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 603                         DEBUGP("nf_conntrack_in: related packet for %p\n", ct);
 604                         *ctinfo = IP_CT_RELATED;
 605                 } else {
 606                         DEBUGP("nf_conntrack_in: new packet for %p\n", ct);
 607                         *ctinfo = IP_CT_NEW;
 608                 }
 609                 *set_reply = 0;
 610         }
 611         skb->nfct = &ct->ct_general;
 612         skb->nfctinfo = *ctinfo;
 613         return ct;
 614 }
 615
 616 unsigned int
 617 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb)
 618 {
 619         struct nf_conn *ct;
 620         enum ip_conntrack_info ctinfo;
 621         struct nf_conntrack_l3proto *l3proto;
 622         struct nf_conntrack_l4proto *l4proto;
 623         unsigned int dataoff;
 624         u_int8_t protonum;
 625         int set_reply = 0;
 626         int ret;
 627
 628         /* Previously seen (loopback or untracked)?  Ignore. */
 629         if ((*pskb)->nfct) {
 630                 NF_CT_STAT_INC_ATOMIC(ignore);
 631                 return NF_ACCEPT;
 632         }
 633
 634         /* rcu_read_lock()ed by nf_hook_slow */
 635         l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
 636
 637         if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) {
 638                 DEBUGP("not prepared to track yet or error occured\n");
 639                 return -ret;
 640         }
 641
 642         l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
 643
 644         /* It may be an special packet, error, unclean...
 645          * inverse of the return code tells to the netfilter
 646          * core what to do with the packet. */
 647         if (l4proto->error != NULL &&
 648             (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
 649                 NF_CT_STAT_INC_ATOMIC(error);
 650                 NF_CT_STAT_INC_ATOMIC(invalid);
 651                 return -ret;
 652         }
 653
 654         ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto,
 655                                &set_reply, &ctinfo);
 656         if (!ct) {
 657                 /* Not valid part of a connection */
 658                 NF_CT_STAT_INC_ATOMIC(invalid);
 659                 return NF_ACCEPT;
 660         }
 661
 662         if (IS_ERR(ct)) {
 663                 /* Too stressed to deal. */
 664                 NF_CT_STAT_INC_ATOMIC(drop);
 665                 return NF_DROP;
 666         }
 667
 668         NF_CT_ASSERT((*pskb)->nfct);
 669
 670         ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum);
 671         if (ret < 0) {
 672                 /* Invalid: inverse of the return code tells
 673                  * the netfilter core what to do */
 674                 DEBUGP("nf_conntrack_in: Can't track with proto module\n");
 675                 nf_conntrack_put((*pskb)->nfct);
 676                 (*pskb)->nfct = NULL;
 677                 NF_CT_STAT_INC_ATOMIC(invalid);
 678                 return -ret;
 679         }
 680
 681         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 682                 nf_conntrack_event_cache(IPCT_STATUS, *pskb);
 683
 684         return ret;
 685 }
 686 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 687
 688 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 689                          const struct nf_conntrack_tuple *orig)
 690 {
 691         int ret;
 692
 693         rcu_read_lock();
 694         ret = nf_ct_invert_tuple(inverse, orig,
 695                                  __nf_ct_l3proto_find(orig->src.l3num),
 696                                  __nf_ct_l4proto_find(orig->src.l3num,
 697                                                       orig->dst.protonum));
 698         rcu_read_unlock();
 699         return ret;
 700 }
 701 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 702
 703 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 704    implicitly racy: see __nf_conntrack_confirm */
 705 void nf_conntrack_alter_reply(struct nf_conn *ct,
 706                               const struct nf_conntrack_tuple *newreply)
 707 {
 708         struct nf_conn_help *help = nfct_help(ct);
 709         struct nf_conntrack_helper *helper;
 710
 711         write_lock_bh(&nf_conntrack_lock);
 712         /* Should be unconfirmed, so not in hash table yet */
 713         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 714
 715         DEBUGP("Altering reply tuple of %p to ", ct);
 716         NF_CT_DUMP_TUPLE(newreply);
 717
 718         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 719         if (ct->master || (help && help->expecting != 0))
 720                 goto out;
 721
 722         helper = __nf_ct_helper_find(newreply);
 723         if (helper == NULL) {
 724                 if (help)
 725                         rcu_assign_pointer(help->helper, NULL);
 726                 goto out;
 727         }
 728
 729         if (help == NULL) {
 730                 help = nf_ct_ext_add(ct, NF_CT_EXT_HELPER, GFP_ATOMIC);
 731                 if (help == NULL) {
 732                         DEBUGP("failed to add helper extension area");
 733                         goto out;
 734                 }
 735         } else {
 736                 memset(&help->help, 0, sizeof(help->help));
 737         }
 738
 739         rcu_assign_pointer(help->helper, helper);
 740 out:
 741         write_unlock_bh(&nf_conntrack_lock);
 742 }
 743 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 744
 745 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 746 void __nf_ct_refresh_acct(struct nf_conn *ct,
 747                           enum ip_conntrack_info ctinfo,
 748                           const struct sk_buff *skb,
 749                           unsigned long extra_jiffies,
 750                           int do_acct)
 751 {
 752         int event = 0;
 753
 754         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 755         NF_CT_ASSERT(skb);
 756
 757         write_lock_bh(&nf_conntrack_lock);
 758
 759         /* Only update if this is not a fixed timeout */
 760         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) {
 761                 write_unlock_bh(&nf_conntrack_lock);
 762                 return;
 763         }
 764
 765         /* If not in hash table, timer will not be active yet */
 766         if (!nf_ct_is_confirmed(ct)) {
 767                 ct->timeout.expires = extra_jiffies;
 768                 event = IPCT_REFRESH;
 769         } else {
 770                 unsigned long newtime = jiffies + extra_jiffies;
 771
 772                 /* Only update the timeout if the new timeout is at least
 773                    HZ jiffies from the old timeout. Need del_timer for race
 774                    avoidance (may already be dying). */
 775                 if (newtime - ct->timeout.expires >= HZ
 776                     && del_timer(&ct->timeout)) {
 777                         ct->timeout.expires = newtime;
 778                         add_timer(&ct->timeout);
 779                         event = IPCT_REFRESH;
 780                 }
 781         }
 782
 783 #ifdef CONFIG_NF_CT_ACCT
 784         if (do_acct) {
 785                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
 786                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
 787                         skb->len - skb_network_offset(skb);
 788
 789                 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
 790                     || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
 791                         event |= IPCT_COUNTER_FILLING;
 792         }
 793 #endif
 794
 795         write_unlock_bh(&nf_conntrack_lock);
 796
 797         /* must be unlocked when calling event cache */
 798         if (event)
 799                 nf_conntrack_event_cache(event, skb);
 800 }
 801 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 802
 803 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 804
 805 #include <linux/netfilter/nfnetlink.h>
 806 #include <linux/netfilter/nfnetlink_conntrack.h>
 807 #include <linux/mutex.h>
 808
 809
 810 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 811  * in ip_conntrack_core, since we don't want the protocols to autoload
 812  * or depend on ctnetlink */
 813 int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb,
 814                                const struct nf_conntrack_tuple *tuple)
 815 {
 816         NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t),
 817                 &tuple->src.u.tcp.port);
 818         NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t),
 819                 &tuple->dst.u.tcp.port);
 820         return 0;
 821
 822 nfattr_failure:
 823         return -1;
 824 }
 825 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr);
 826
 827 static const size_t cta_min_proto[CTA_PROTO_MAX] = {
 828         [CTA_PROTO_SRC_PORT-1]  = sizeof(u_int16_t),
 829         [CTA_PROTO_DST_PORT-1]  = sizeof(u_int16_t)
 830 };
 831
 832 int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[],
 833                                struct nf_conntrack_tuple *t)
 834 {
 835         if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1])
 836                 return -EINVAL;
 837
 838         if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
 839                 return -EINVAL;
 840
 841         t->src.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]);
 842         t->dst.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]);
 843
 844         return 0;
 845 }
 846 EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple);
 847 #endif
 848
 849 /* Used by ipt_REJECT and ip6t_REJECT. */
 850 void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 851 {
 852         struct nf_conn *ct;
 853         enum ip_conntrack_info ctinfo;
 854
 855         /* This ICMP is in reverse direction to the packet which caused it */
 856         ct = nf_ct_get(skb, &ctinfo);
 857         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 858                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 859         else
 860                 ctinfo = IP_CT_RELATED;
 861
 862         /* Attach to new skbuff, and increment count */
 863         nskb->nfct = &ct->ct_general;
 864         nskb->nfctinfo = ctinfo;
 865         nf_conntrack_get(nskb->nfct);
 866 }
 867 EXPORT_SYMBOL_GPL(__nf_conntrack_attach);
 868
 869 static inline int
 870 do_iter(const struct nf_conntrack_tuple_hash *i,
 871         int (*iter)(struct nf_conn *i, void *data),
 872         void *data)
 873 {
 874         return iter(nf_ct_tuplehash_to_ctrack(i), data);
 875 }
 876
 877 /* Bring out ya dead! */
 878 static struct nf_conn *
 879 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
 880                 void *data, unsigned int *bucket)
 881 {
 882         struct nf_conntrack_tuple_hash *h;
 883         struct nf_conn *ct;
 884
 885         write_lock_bh(&nf_conntrack_lock);
 886         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 887                 list_for_each_entry(h, &nf_conntrack_hash[*bucket], list) {
 888                         ct = nf_ct_tuplehash_to_ctrack(h);
 889                         if (iter(ct, data))
 890                                 goto found;
 891                 }
 892         }
 893         list_for_each_entry(h, &unconfirmed, list) {
 894                 ct = nf_ct_tuplehash_to_ctrack(h);
 895                 if (iter(ct, data))
 896                         set_bit(IPS_DYING_BIT, &ct->status);
 897         }
 898         write_unlock_bh(&nf_conntrack_lock);
 899         return NULL;
 900 found:
 901         atomic_inc(&ct->ct_general.use);
 902         write_unlock_bh(&nf_conntrack_lock);
 903         return ct;
 904 }
 905
 906 void
 907 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
 908 {
 909         struct nf_conn *ct;
 910         unsigned int bucket = 0;
 911
 912         while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
 913                 /* Time to push up daises... */
 914                 if (del_timer(&ct->timeout))
 915                         death_by_timeout((unsigned long)ct);
 916                 /* ... else the timer will get him soon. */
 917
 918                 nf_ct_put(ct);
 919         }
 920 }
 921 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
 922
 923 static int kill_all(struct nf_conn *i, void *data)
 924 {
 925         return 1;
 926 }
 927
 928 static void free_conntrack_hash(struct list_head *hash, int vmalloced, int size)
 929 {
 930         if (vmalloced)
 931                 vfree(hash);
 932         else
 933                 free_pages((unsigned long)hash,
 934                            get_order(sizeof(struct list_head) * size));
 935 }
 936
 937 void nf_conntrack_flush(void)
 938 {
 939         nf_ct_iterate_cleanup(kill_all, NULL);
 940 }
 941 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
 942
 943 /* Mishearing the voices in his head, our hero wonders how he's
 944    supposed to kill the mall. */
 945 void nf_conntrack_cleanup(void)
 946 {
 947         rcu_assign_pointer(ip_ct_attach, NULL);
 948
 949         /* This makes sure all current packets have passed through
 950            netfilter framework.  Roll on, two-stage module
 951            delete... */
 952         synchronize_net();
 953
 954         nf_ct_event_cache_flush();
 955  i_see_dead_people:
 956         nf_conntrack_flush();
 957         if (atomic_read(&nf_conntrack_count) != 0) {
 958                 schedule();
 959                 goto i_see_dead_people;
 960         }
 961         /* wait until all references to nf_conntrack_untracked are dropped */
 962         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
 963                 schedule();
 964
 965         rcu_assign_pointer(nf_ct_destroy, NULL);
 966
 967         kmem_cache_destroy(nf_conntrack_cachep);
 968         kmem_cache_destroy(nf_conntrack_expect_cachep);
 969         free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
 970                             nf_conntrack_htable_size);
 971
 972         nf_conntrack_proto_fini();
 973         nf_conntrack_helper_fini();
 974 }
 975
 976 static struct list_head *alloc_hashtable(int size, int *vmalloced)
 977 {
 978         struct list_head *hash;
 979         unsigned int i;
 980
 981         *vmalloced = 0;
 982         hash = (void*)__get_free_pages(GFP_KERNEL,
 983                                        get_order(sizeof(struct list_head)
 984                                                  * size));
 985         if (!hash) {
 986                 *vmalloced = 1;
 987                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
 988                 hash = vmalloc(sizeof(struct list_head) * size);
 989         }
 990
 991         if (hash)
 992                 for (i = 0; i < size; i++)
 993                         INIT_LIST_HEAD(&hash[i]);
 994
 995         return hash;
 996 }
 997
 998 int set_hashsize(const char *val, struct kernel_param *kp)
 999 {
1000         int i, bucket, hashsize, vmalloced;
1001         int old_vmalloced, old_size;
1002         int rnd;
1003         struct list_head *hash, *old_hash;
1004         struct nf_conntrack_tuple_hash *h;
1005
1006         /* On boot, we can set this without any fancy locking. */
1007         if (!nf_conntrack_htable_size)
1008                 return param_set_uint(val, kp);
1009
1010         hashsize = simple_strtol(val, NULL, 0);
1011         if (!hashsize)
1012                 return -EINVAL;
1013
1014         hash = alloc_hashtable(hashsize, &vmalloced);
1015         if (!hash)
1016                 return -ENOMEM;
1017
1018         /* We have to rehahs for the new table anyway, so we also can
1019          * use a newrandom seed */
1020         get_random_bytes(&rnd, 4);
1021
1022         write_lock_bh(&nf_conntrack_lock);
1023         for (i = 0; i < nf_conntrack_htable_size; i++) {
1024                 while (!list_empty(&nf_conntrack_hash[i])) {
1025                         h = list_entry(nf_conntrack_hash[i].next,
1026                                        struct nf_conntrack_tuple_hash, list);
1027                         list_del(&h->list);
1028                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1029                         list_add_tail(&h->list, &hash[bucket]);
1030                 }
1031         }
1032         old_size = nf_conntrack_htable_size;
1033         old_vmalloced = nf_conntrack_vmalloc;
1034         old_hash = nf_conntrack_hash;
1035
1036         nf_conntrack_htable_size = hashsize;
1037         nf_conntrack_vmalloc = vmalloced;
1038         nf_conntrack_hash = hash;
1039         nf_conntrack_hash_rnd = rnd;
1040         write_unlock_bh(&nf_conntrack_lock);
1041
1042         free_conntrack_hash(old_hash, old_vmalloced, old_size);
1043         return 0;
1044 }
1045
1046 module_param_call(hashsize, set_hashsize, param_get_uint,
1047                   &nf_conntrack_htable_size, 0600);
1048
1049 int __init nf_conntrack_init(void)
1050 {
1051         int ret;
1052
1053         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1054          * machine has 256 buckets.  >= 1GB machines have 8192 buckets. */
1055         if (!nf_conntrack_htable_size) {
1056                 nf_conntrack_htable_size
1057                         = (((num_physpages << PAGE_SHIFT) / 16384)
1058                            / sizeof(struct list_head));
1059                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1060                         nf_conntrack_htable_size = 8192;
1061                 if (nf_conntrack_htable_size < 16)
1062                         nf_conntrack_htable_size = 16;
1063         }
1064         nf_conntrack_max = 8 * nf_conntrack_htable_size;
1065
1066         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1067                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1068                nf_conntrack_max);
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_cachep = kmem_cache_create("nf_conntrack",
1078                                                 sizeof(struct nf_conn),
1079                                                 0, 0, NULL, NULL);
1080         if (!nf_conntrack_cachep) {
1081                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1082                 goto err_free_hash;
1083         }
1084
1085         nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect",
1086                                         sizeof(struct nf_conntrack_expect),
1087                                         0, 0, NULL, NULL);
1088         if (!nf_conntrack_expect_cachep) {
1089                 printk(KERN_ERR "Unable to create nf_expect slab cache\n");
1090                 goto err_free_conntrack_slab;
1091         }
1092
1093         ret = nf_conntrack_proto_init();
1094         if (ret < 0)
1095                 goto out_free_expect_slab;
1096
1097         ret = nf_conntrack_helper_init();
1098         if (ret < 0)
1099                 goto out_fini_proto;
1100
1101         /* For use by REJECT target */
1102         rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach);
1103         rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1104
1105         /* Set up fake conntrack:
1106             - to never be deleted, not in any hashes */
1107         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1108         /*  - and look it like as a confirmed connection */
1109         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1110
1111         return ret;
1112
1113 out_fini_proto:
1114         nf_conntrack_proto_fini();
1115 out_free_expect_slab:
1116         kmem_cache_destroy(nf_conntrack_expect_cachep);
1117 err_free_conntrack_slab:
1118         kmem_cache_destroy(nf_conntrack_cachep);
1119 err_free_hash:
1120         free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
1121                             nf_conntrack_htable_size);
1122 err_out:
1123         return -ENOMEM;
1124 }