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