err.no Git - 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 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 list_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 LIST_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         list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
 146         list_del(&ct->tuplehash[IP_CT_DIR_REPLY].list);
 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(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list));
 188                 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
 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         unsigned int hash = hash_conntrack(tuple);
 230
 231         list_for_each_entry(h, &nf_conntrack_hash[hash], list) {
 232                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
 233                     nf_ct_tuple_equal(tuple, &h->tuple)) {
 234                         NF_CT_STAT_INC(found);
 235                         return h;
 236                 }
 237                 NF_CT_STAT_INC(searched);
 238         }
 239
 240         return NULL;
 241 }
 242 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 243
 244 /* Find a connection corresponding to a tuple. */
 245 struct nf_conntrack_tuple_hash *
 246 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple,
 247                       const struct nf_conn *ignored_conntrack)
 248 {
 249         struct nf_conntrack_tuple_hash *h;
 250
 251         read_lock_bh(&nf_conntrack_lock);
 252         h = __nf_conntrack_find(tuple, ignored_conntrack);
 253         if (h)
 254                 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
 255         read_unlock_bh(&nf_conntrack_lock);
 256
 257         return h;
 258 }
 259 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 260
 261 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 262                                        unsigned int hash,
 263                                        unsigned int repl_hash)
 264 {
 265         ct->id = ++nf_conntrack_next_id;
 266         list_add(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list,
 267                  &nf_conntrack_hash[hash]);
 268         list_add(&ct->tuplehash[IP_CT_DIR_REPLY].list,
 269                  &nf_conntrack_hash[repl_hash]);
 270 }
 271
 272 void nf_conntrack_hash_insert(struct nf_conn *ct)
 273 {
 274         unsigned int hash, repl_hash;
 275
 276         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 277         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 278
 279         write_lock_bh(&nf_conntrack_lock);
 280         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 281         write_unlock_bh(&nf_conntrack_lock);
 282 }
 283 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 284
 285 /* Confirm a connection given skb; places it in hash table */
 286 int
 287 __nf_conntrack_confirm(struct sk_buff **pskb)
 288 {
 289         unsigned int hash, repl_hash;
 290         struct nf_conntrack_tuple_hash *h;
 291         struct nf_conn *ct;
 292         struct nf_conn_help *help;
 293         enum ip_conntrack_info ctinfo;
 294
 295         ct = nf_ct_get(*pskb, &ctinfo);
 296
 297         /* ipt_REJECT uses nf_conntrack_attach to attach related
 298            ICMP/TCP RST packets in other direction.  Actual packet
 299            which created connection will be IP_CT_NEW or for an
 300            expected connection, IP_CT_RELATED. */
 301         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 302                 return NF_ACCEPT;
 303
 304         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 305         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 306
 307         /* We're not in hash table, and we refuse to set up related
 308            connections for unconfirmed conns.  But packet copies and
 309            REJECT will give spurious warnings here. */
 310         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 311
 312         /* No external references means noone else could have
 313            confirmed us. */
 314         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 315         DEBUGP("Confirming conntrack %p\n", ct);
 316
 317         write_lock_bh(&nf_conntrack_lock);
 318
 319         /* See if there's one in the list already, including reverse:
 320            NAT could have grabbed it without realizing, since we're
 321            not in the hash.  If there is, we lost race. */
 322         list_for_each_entry(h, &nf_conntrack_hash[hash], list)
 323                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 324                                       &h->tuple))
 325                         goto out;
 326         list_for_each_entry(h, &nf_conntrack_hash[repl_hash], list)
 327                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 328                                       &h->tuple))
 329                         goto out;
 330
 331         /* Remove from unconfirmed list */
 332         list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
 333
 334         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 335         /* Timer relative to confirmation time, not original
 336            setting time, otherwise we'd get timer wrap in
 337            weird delay cases. */
 338         ct->timeout.expires += jiffies;
 339         add_timer(&ct->timeout);
 340         atomic_inc(&ct->ct_general.use);
 341         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 342         NF_CT_STAT_INC(insert);
 343         write_unlock_bh(&nf_conntrack_lock);
 344         help = nfct_help(ct);
 345         if (help && help->helper)
 346                 nf_conntrack_event_cache(IPCT_HELPER, *pskb);
 347 #ifdef CONFIG_NF_NAT_NEEDED
 348         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
 349             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
 350                 nf_conntrack_event_cache(IPCT_NATINFO, *pskb);
 351 #endif
 352         nf_conntrack_event_cache(master_ct(ct) ?
 353                                  IPCT_RELATED : IPCT_NEW, *pskb);
 354         return NF_ACCEPT;
 355
 356 out:
 357         NF_CT_STAT_INC(insert_failed);
 358         write_unlock_bh(&nf_conntrack_lock);
 359         return NF_DROP;
 360 }
 361 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 362
 363 /* Returns true if a connection correspondings to the tuple (required
 364    for NAT). */
 365 int
 366 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 367                          const struct nf_conn *ignored_conntrack)
 368 {
 369         struct nf_conntrack_tuple_hash *h;
 370
 371         read_lock_bh(&nf_conntrack_lock);
 372         h = __nf_conntrack_find(tuple, ignored_conntrack);
 373         read_unlock_bh(&nf_conntrack_lock);
 374
 375         return h != NULL;
 376 }
 377 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 378
 379 /* There's a small race here where we may free a just-assured
 380    connection.  Too bad: we're in trouble anyway. */
 381 static int early_drop(struct list_head *chain)
 382 {
 383         /* Traverse backwards: gives us oldest, which is roughly LRU */
 384         struct nf_conntrack_tuple_hash *h;
 385         struct nf_conn *ct = NULL, *tmp;
 386         int dropped = 0;
 387
 388         read_lock_bh(&nf_conntrack_lock);
 389         list_for_each_entry_reverse(h, chain, list) {
 390                 tmp = nf_ct_tuplehash_to_ctrack(h);
 391                 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) {
 392                         ct = tmp;
 393                         atomic_inc(&ct->ct_general.use);
 394                         break;
 395                 }
 396         }
 397         read_unlock_bh(&nf_conntrack_lock);
 398
 399         if (!ct)
 400                 return dropped;
 401
 402         if (del_timer(&ct->timeout)) {
 403                 death_by_timeout((unsigned long)ct);
 404                 dropped = 1;
 405                 NF_CT_STAT_INC_ATOMIC(early_drop);
 406         }
 407         nf_ct_put(ct);
 408         return dropped;
 409 }
 410
 411 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
 412                                    const struct nf_conntrack_tuple *repl)
 413 {
 414         struct nf_conn *conntrack = NULL;
 415
 416         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 417                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
 418                 nf_conntrack_hash_rnd_initted = 1;
 419         }
 420
 421         /* We don't want any race condition at early drop stage */
 422         atomic_inc(&nf_conntrack_count);
 423
 424         if (nf_conntrack_max
 425             && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
 426                 unsigned int hash = hash_conntrack(orig);
 427                 /* Try dropping from this hash chain. */
 428                 if (!early_drop(&nf_conntrack_hash[hash])) {
 429                         atomic_dec(&nf_conntrack_count);
 430                         if (net_ratelimit())
 431                                 printk(KERN_WARNING
 432                                        "nf_conntrack: table full, dropping"
 433                                        " packet.\n");
 434                         return ERR_PTR(-ENOMEM);
 435                 }
 436         }
 437
 438         conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
 439         if (conntrack == NULL) {
 440                 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack.\n");
 441                 atomic_dec(&nf_conntrack_count);
 442                 return ERR_PTR(-ENOMEM);
 443         }
 444
 445         atomic_set(&conntrack->ct_general.use, 1);
 446         conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 447         conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 448         /* Don't set timer yet: wait for confirmation */
 449         setup_timer(&conntrack->timeout, death_by_timeout,
 450                     (unsigned long)conntrack);
 451
 452         return conntrack;
 453 }
 454 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 455
 456 void nf_conntrack_free(struct nf_conn *conntrack)
 457 {
 458         nf_ct_ext_free(conntrack);
 459         kmem_cache_free(nf_conntrack_cachep, conntrack);
 460         atomic_dec(&nf_conntrack_count);
 461 }
 462 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 463
 464 /* Allocate a new conntrack: we return -ENOMEM if classification
 465    failed due to stress.  Otherwise it really is unclassifiable. */
 466 static struct nf_conntrack_tuple_hash *
 467 init_conntrack(const struct nf_conntrack_tuple *tuple,
 468                struct nf_conntrack_l3proto *l3proto,
 469                struct nf_conntrack_l4proto *l4proto,
 470                struct sk_buff *skb,
 471                unsigned int dataoff)
 472 {
 473         struct nf_conn *conntrack;
 474         struct nf_conn_help *help;
 475         struct nf_conntrack_tuple repl_tuple;
 476         struct nf_conntrack_expect *exp;
 477
 478         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 479                 DEBUGP("Can't invert tuple.\n");
 480                 return NULL;
 481         }
 482
 483         conntrack = nf_conntrack_alloc(tuple, &repl_tuple);
 484         if (conntrack == NULL || IS_ERR(conntrack)) {
 485                 DEBUGP("Can't allocate conntrack.\n");
 486                 return (struct nf_conntrack_tuple_hash *)conntrack;
 487         }
 488
 489         if (!l4proto->new(conntrack, skb, dataoff)) {
 490                 nf_conntrack_free(conntrack);
 491                 DEBUGP("init conntrack: can't track with proto module\n");
 492                 return NULL;
 493         }
 494
 495         write_lock_bh(&nf_conntrack_lock);
 496         exp = find_expectation(tuple);
 497         if (exp) {
 498                 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
 499                         conntrack, exp);
 500                 /* Welcome, Mr. Bond.  We've been expecting you... */
 501                 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
 502                 conntrack->master = exp->master;
 503                 if (exp->helper) {
 504                         help = nf_ct_ext_add(conntrack, NF_CT_EXT_HELPER,
 505                                              GFP_ATOMIC);
 506                         if (help)
 507                                 rcu_assign_pointer(help->helper, exp->helper);
 508                         else
 509                                 DEBUGP("failed to add helper extension area");
 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_ext_add(conntrack, NF_CT_EXT_HELPER,
 526                                              GFP_ATOMIC);
 527                         if (help)
 528                                 /* not in hash table yet, so not strictly
 529                                    necessary */
 530                                 rcu_assign_pointer(help->helper, helper);
 531                         else
 532                                 DEBUGP("failed to add helper extension area");
 533                 }
 534                 NF_CT_STAT_INC(new);
 535         }
 536
 537         /* Overload tuple linked list to put us in unconfirmed list. */
 538         list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed);
 539
 540         write_unlock_bh(&nf_conntrack_lock);
 541
 542         if (exp) {
 543                 if (exp->expectfn)
 544                         exp->expectfn(conntrack, exp);
 545                 nf_conntrack_expect_put(exp);
 546         }
 547
 548         return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
 549 }
 550
 551 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 552 static inline struct nf_conn *
 553 resolve_normal_ct(struct sk_buff *skb,
 554                   unsigned int dataoff,
 555                   u_int16_t l3num,
 556                   u_int8_t protonum,
 557                   struct nf_conntrack_l3proto *l3proto,
 558                   struct nf_conntrack_l4proto *l4proto,
 559                   int *set_reply,
 560                   enum ip_conntrack_info *ctinfo)
 561 {
 562         struct nf_conntrack_tuple tuple;
 563         struct nf_conntrack_tuple_hash *h;
 564         struct nf_conn *ct;
 565
 566         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 567                              dataoff, l3num, protonum, &tuple, l3proto,
 568                              l4proto)) {
 569                 DEBUGP("resolve_normal_ct: Can't get tuple\n");
 570                 return NULL;
 571         }
 572
 573         /* look for tuple match */
 574         h = nf_conntrack_find_get(&tuple, NULL);
 575         if (!h) {
 576                 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
 577                 if (!h)
 578                         return NULL;
 579                 if (IS_ERR(h))
 580                         return (void *)h;
 581         }
 582         ct = nf_ct_tuplehash_to_ctrack(h);
 583
 584         /* It exists; we have (non-exclusive) reference. */
 585         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 586                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 587                 /* Please set reply bit if this packet OK */
 588                 *set_reply = 1;
 589         } else {
 590                 /* Once we've had two way comms, always ESTABLISHED. */
 591                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 592                         DEBUGP("nf_conntrack_in: normal packet for %p\n", ct);
 593                         *ctinfo = IP_CT_ESTABLISHED;
 594                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 595                         DEBUGP("nf_conntrack_in: related packet for %p\n", ct);
 596                         *ctinfo = IP_CT_RELATED;
 597                 } else {
 598                         DEBUGP("nf_conntrack_in: new packet for %p\n", ct);
 599                         *ctinfo = IP_CT_NEW;
 600                 }
 601                 *set_reply = 0;
 602         }
 603         skb->nfct = &ct->ct_general;
 604         skb->nfctinfo = *ctinfo;
 605         return ct;
 606 }
 607
 608 unsigned int
 609 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb)
 610 {
 611         struct nf_conn *ct;
 612         enum ip_conntrack_info ctinfo;
 613         struct nf_conntrack_l3proto *l3proto;
 614         struct nf_conntrack_l4proto *l4proto;
 615         unsigned int dataoff;
 616         u_int8_t protonum;
 617         int set_reply = 0;
 618         int ret;
 619
 620         /* Previously seen (loopback or untracked)?  Ignore. */
 621         if ((*pskb)->nfct) {
 622                 NF_CT_STAT_INC_ATOMIC(ignore);
 623                 return NF_ACCEPT;
 624         }
 625
 626         /* rcu_read_lock()ed by nf_hook_slow */
 627         l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
 628
 629         if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) {
 630                 DEBUGP("not prepared to track yet or error occured\n");
 631                 return -ret;
 632         }
 633
 634         l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
 635
 636         /* It may be an special packet, error, unclean...
 637          * inverse of the return code tells to the netfilter
 638          * core what to do with the packet. */
 639         if (l4proto->error != NULL &&
 640             (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
 641                 NF_CT_STAT_INC_ATOMIC(error);
 642                 NF_CT_STAT_INC_ATOMIC(invalid);
 643                 return -ret;
 644         }
 645
 646         ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto,
 647                                &set_reply, &ctinfo);
 648         if (!ct) {
 649                 /* Not valid part of a connection */
 650                 NF_CT_STAT_INC_ATOMIC(invalid);
 651                 return NF_ACCEPT;
 652         }
 653
 654         if (IS_ERR(ct)) {
 655                 /* Too stressed to deal. */
 656                 NF_CT_STAT_INC_ATOMIC(drop);
 657                 return NF_DROP;
 658         }
 659
 660         NF_CT_ASSERT((*pskb)->nfct);
 661
 662         ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum);
 663         if (ret < 0) {
 664                 /* Invalid: inverse of the return code tells
 665                  * the netfilter core what to do */
 666                 DEBUGP("nf_conntrack_in: Can't track with proto module\n");
 667                 nf_conntrack_put((*pskb)->nfct);
 668                 (*pskb)->nfct = NULL;
 669                 NF_CT_STAT_INC_ATOMIC(invalid);
 670                 return -ret;
 671         }
 672
 673         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 674                 nf_conntrack_event_cache(IPCT_STATUS, *pskb);
 675
 676         return ret;
 677 }
 678 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 679
 680 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 681                          const struct nf_conntrack_tuple *orig)
 682 {
 683         int ret;
 684
 685         rcu_read_lock();
 686         ret = nf_ct_invert_tuple(inverse, orig,
 687                                  __nf_ct_l3proto_find(orig->src.l3num),
 688                                  __nf_ct_l4proto_find(orig->src.l3num,
 689                                                       orig->dst.protonum));
 690         rcu_read_unlock();
 691         return ret;
 692 }
 693 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 694
 695 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 696    implicitly racy: see __nf_conntrack_confirm */
 697 void nf_conntrack_alter_reply(struct nf_conn *ct,
 698                               const struct nf_conntrack_tuple *newreply)
 699 {
 700         struct nf_conn_help *help = nfct_help(ct);
 701         struct nf_conntrack_helper *helper;
 702
 703         write_lock_bh(&nf_conntrack_lock);
 704         /* Should be unconfirmed, so not in hash table yet */
 705         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 706
 707         DEBUGP("Altering reply tuple of %p to ", ct);
 708         NF_CT_DUMP_TUPLE(newreply);
 709
 710         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 711         if (ct->master || (help && help->expecting != 0))
 712                 goto out;
 713
 714         helper = __nf_ct_helper_find(newreply);
 715         if (helper == NULL) {
 716                 if (help)
 717                         rcu_assign_pointer(help->helper, NULL);
 718                 goto out;
 719         }
 720
 721         if (help == NULL) {
 722                 help = nf_ct_ext_add(ct, NF_CT_EXT_HELPER, GFP_ATOMIC);
 723                 if (help == NULL) {
 724                         DEBUGP("failed to add helper extension area");
 725                         goto out;
 726                 }
 727         } else {
 728                 memset(&help->help, 0, sizeof(help->help));
 729         }
 730
 731         rcu_assign_pointer(help->helper, helper);
 732 out:
 733         write_unlock_bh(&nf_conntrack_lock);
 734 }
 735 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 736
 737 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 738 void __nf_ct_refresh_acct(struct nf_conn *ct,
 739                           enum ip_conntrack_info ctinfo,
 740                           const struct sk_buff *skb,
 741                           unsigned long extra_jiffies,
 742                           int do_acct)
 743 {
 744         int event = 0;
 745
 746         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 747         NF_CT_ASSERT(skb);
 748
 749         write_lock_bh(&nf_conntrack_lock);
 750
 751         /* Only update if this is not a fixed timeout */
 752         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) {
 753                 write_unlock_bh(&nf_conntrack_lock);
 754                 return;
 755         }
 756
 757         /* If not in hash table, timer will not be active yet */
 758         if (!nf_ct_is_confirmed(ct)) {
 759                 ct->timeout.expires = extra_jiffies;
 760                 event = IPCT_REFRESH;
 761         } else {
 762                 unsigned long newtime = jiffies + extra_jiffies;
 763
 764                 /* Only update the timeout if the new timeout is at least
 765                    HZ jiffies from the old timeout. Need del_timer for race
 766                    avoidance (may already be dying). */
 767                 if (newtime - ct->timeout.expires >= HZ
 768                     && del_timer(&ct->timeout)) {
 769                         ct->timeout.expires = newtime;
 770                         add_timer(&ct->timeout);
 771                         event = IPCT_REFRESH;
 772                 }
 773         }
 774
 775 #ifdef CONFIG_NF_CT_ACCT
 776         if (do_acct) {
 777                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
 778                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
 779                         skb->len - skb_network_offset(skb);
 780
 781                 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
 782                     || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
 783                         event |= IPCT_COUNTER_FILLING;
 784         }
 785 #endif
 786
 787         write_unlock_bh(&nf_conntrack_lock);
 788
 789         /* must be unlocked when calling event cache */
 790         if (event)
 791                 nf_conntrack_event_cache(event, skb);
 792 }
 793 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 794
 795 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 796
 797 #include <linux/netfilter/nfnetlink.h>
 798 #include <linux/netfilter/nfnetlink_conntrack.h>
 799 #include <linux/mutex.h>
 800
 801
 802 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 803  * in ip_conntrack_core, since we don't want the protocols to autoload
 804  * or depend on ctnetlink */
 805 int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb,
 806                                const struct nf_conntrack_tuple *tuple)
 807 {
 808         NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t),
 809                 &tuple->src.u.tcp.port);
 810         NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t),
 811                 &tuple->dst.u.tcp.port);
 812         return 0;
 813
 814 nfattr_failure:
 815         return -1;
 816 }
 817 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr);
 818
 819 static const size_t cta_min_proto[CTA_PROTO_MAX] = {
 820         [CTA_PROTO_SRC_PORT-1]  = sizeof(u_int16_t),
 821         [CTA_PROTO_DST_PORT-1]  = sizeof(u_int16_t)
 822 };
 823
 824 int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[],
 825                                struct nf_conntrack_tuple *t)
 826 {
 827         if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1])
 828                 return -EINVAL;
 829
 830         if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
 831                 return -EINVAL;
 832
 833         t->src.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]);
 834         t->dst.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]);
 835
 836         return 0;
 837 }
 838 EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple);
 839 #endif
 840
 841 /* Used by ipt_REJECT and ip6t_REJECT. */
 842 void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 843 {
 844         struct nf_conn *ct;
 845         enum ip_conntrack_info ctinfo;
 846
 847         /* This ICMP is in reverse direction to the packet which caused it */
 848         ct = nf_ct_get(skb, &ctinfo);
 849         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 850                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 851         else
 852                 ctinfo = IP_CT_RELATED;
 853
 854         /* Attach to new skbuff, and increment count */
 855         nskb->nfct = &ct->ct_general;
 856         nskb->nfctinfo = ctinfo;
 857         nf_conntrack_get(nskb->nfct);
 858 }
 859 EXPORT_SYMBOL_GPL(__nf_conntrack_attach);
 860
 861 static inline int
 862 do_iter(const struct nf_conntrack_tuple_hash *i,
 863         int (*iter)(struct nf_conn *i, void *data),
 864         void *data)
 865 {
 866         return iter(nf_ct_tuplehash_to_ctrack(i), data);
 867 }
 868
 869 /* Bring out ya dead! */
 870 static struct nf_conn *
 871 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
 872                 void *data, unsigned int *bucket)
 873 {
 874         struct nf_conntrack_tuple_hash *h;
 875         struct nf_conn *ct;
 876
 877         write_lock_bh(&nf_conntrack_lock);
 878         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 879                 list_for_each_entry(h, &nf_conntrack_hash[*bucket], list) {
 880                         ct = nf_ct_tuplehash_to_ctrack(h);
 881                         if (iter(ct, data))
 882                                 goto found;
 883                 }
 884         }
 885         list_for_each_entry(h, &unconfirmed, list) {
 886                 ct = nf_ct_tuplehash_to_ctrack(h);
 887                 if (iter(ct, data))
 888                         set_bit(IPS_DYING_BIT, &ct->status);
 889         }
 890         write_unlock_bh(&nf_conntrack_lock);
 891         return NULL;
 892 found:
 893         atomic_inc(&ct->ct_general.use);
 894         write_unlock_bh(&nf_conntrack_lock);
 895         return ct;
 896 }
 897
 898 void
 899 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
 900 {
 901         struct nf_conn *ct;
 902         unsigned int bucket = 0;
 903
 904         while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
 905                 /* Time to push up daises... */
 906                 if (del_timer(&ct->timeout))
 907                         death_by_timeout((unsigned long)ct);
 908                 /* ... else the timer will get him soon. */
 909
 910                 nf_ct_put(ct);
 911         }
 912 }
 913 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
 914
 915 static int kill_all(struct nf_conn *i, void *data)
 916 {
 917         return 1;
 918 }
 919
 920 static void free_conntrack_hash(struct list_head *hash, int vmalloced, int size)
 921 {
 922         if (vmalloced)
 923                 vfree(hash);
 924         else
 925                 free_pages((unsigned long)hash,
 926                            get_order(sizeof(struct list_head) * size));
 927 }
 928
 929 void nf_conntrack_flush(void)
 930 {
 931         nf_ct_iterate_cleanup(kill_all, NULL);
 932 }
 933 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
 934
 935 /* Mishearing the voices in his head, our hero wonders how he's
 936    supposed to kill the mall. */
 937 void nf_conntrack_cleanup(void)
 938 {
 939         rcu_assign_pointer(ip_ct_attach, NULL);
 940
 941         /* This makes sure all current packets have passed through
 942            netfilter framework.  Roll on, two-stage module
 943            delete... */
 944         synchronize_net();
 945
 946         nf_ct_event_cache_flush();
 947  i_see_dead_people:
 948         nf_conntrack_flush();
 949         if (atomic_read(&nf_conntrack_count) != 0) {
 950                 schedule();
 951                 goto i_see_dead_people;
 952         }
 953         /* wait until all references to nf_conntrack_untracked are dropped */
 954         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
 955                 schedule();
 956
 957         rcu_assign_pointer(nf_ct_destroy, NULL);
 958
 959         kmem_cache_destroy(nf_conntrack_cachep);
 960         kmem_cache_destroy(nf_conntrack_expect_cachep);
 961         free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
 962                             nf_conntrack_htable_size);
 963
 964         nf_conntrack_proto_fini();
 965         nf_conntrack_helper_fini();
 966 }
 967
 968 static struct list_head *alloc_hashtable(int *sizep, int *vmalloced)
 969 {
 970         struct list_head *hash;
 971         unsigned int size, i;
 972
 973         *vmalloced = 0;
 974
 975         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct list_head));
 976         hash = (void*)__get_free_pages(GFP_KERNEL,
 977                                        get_order(sizeof(struct list_head)
 978                                                  * size));
 979         if (!hash) {
 980                 *vmalloced = 1;
 981                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
 982                 hash = vmalloc(sizeof(struct list_head) * size);
 983         }
 984
 985         if (hash)
 986                 for (i = 0; i < size; i++)
 987                         INIT_LIST_HEAD(&hash[i]);
 988
 989         return hash;
 990 }
 991
 992 int set_hashsize(const char *val, struct kernel_param *kp)
 993 {
 994         int i, bucket, hashsize, vmalloced;
 995         int old_vmalloced, old_size;
 996         int rnd;
 997         struct list_head *hash, *old_hash;
 998         struct nf_conntrack_tuple_hash *h;
 999
1000         /* On boot, we can set this without any fancy locking. */
1001         if (!nf_conntrack_htable_size)
1002                 return param_set_uint(val, kp);
1003
1004         hashsize = simple_strtol(val, NULL, 0);
1005         if (!hashsize)
1006                 return -EINVAL;
1007
1008         hash = alloc_hashtable(&hashsize, &vmalloced);
1009         if (!hash)
1010                 return -ENOMEM;
1011
1012         /* We have to rehahs for the new table anyway, so we also can
1013          * use a newrandom seed */
1014         get_random_bytes(&rnd, 4);
1015
1016         write_lock_bh(&nf_conntrack_lock);
1017         for (i = 0; i < nf_conntrack_htable_size; i++) {
1018                 while (!list_empty(&nf_conntrack_hash[i])) {
1019                         h = list_entry(nf_conntrack_hash[i].next,
1020                                        struct nf_conntrack_tuple_hash, list);
1021                         list_del(&h->list);
1022                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1023                         list_add_tail(&h->list, &hash[bucket]);
1024                 }
1025         }
1026         old_size = nf_conntrack_htable_size;
1027         old_vmalloced = nf_conntrack_vmalloc;
1028         old_hash = nf_conntrack_hash;
1029
1030         nf_conntrack_htable_size = hashsize;
1031         nf_conntrack_vmalloc = vmalloced;
1032         nf_conntrack_hash = hash;
1033         nf_conntrack_hash_rnd = rnd;
1034         write_unlock_bh(&nf_conntrack_lock);
1035
1036         free_conntrack_hash(old_hash, old_vmalloced, old_size);
1037         return 0;
1038 }
1039
1040 module_param_call(hashsize, set_hashsize, param_get_uint,
1041                   &nf_conntrack_htable_size, 0600);
1042
1043 int __init nf_conntrack_init(void)
1044 {
1045         int ret;
1046
1047         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1048          * machine has 256 buckets.  >= 1GB machines have 8192 buckets. */
1049         if (!nf_conntrack_htable_size) {
1050                 nf_conntrack_htable_size
1051                         = (((num_physpages << PAGE_SHIFT) / 16384)
1052                            / sizeof(struct list_head));
1053                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1054                         nf_conntrack_htable_size = 8192;
1055                 if (nf_conntrack_htable_size < 16)
1056                         nf_conntrack_htable_size = 16;
1057         }
1058         nf_conntrack_hash = alloc_hashtable(&nf_conntrack_htable_size,
1059                                             &nf_conntrack_vmalloc);
1060         if (!nf_conntrack_hash) {
1061                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1062                 goto err_out;
1063         }
1064
1065         nf_conntrack_max = 8 * nf_conntrack_htable_size;
1066
1067         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1068                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1069                nf_conntrack_max);
1070
1071         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1072                                                 sizeof(struct nf_conn),
1073                                                 0, 0, NULL, NULL);
1074         if (!nf_conntrack_cachep) {
1075                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1076                 goto err_free_hash;
1077         }
1078
1079         nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect",
1080                                         sizeof(struct nf_conntrack_expect),
1081                                         0, 0, NULL, NULL);
1082         if (!nf_conntrack_expect_cachep) {
1083                 printk(KERN_ERR "Unable to create nf_expect slab cache\n");
1084                 goto err_free_conntrack_slab;
1085         }
1086
1087         ret = nf_conntrack_proto_init();
1088         if (ret < 0)
1089                 goto out_free_expect_slab;
1090
1091         ret = nf_conntrack_helper_init();
1092         if (ret < 0)
1093                 goto out_fini_proto;
1094
1095         /* For use by REJECT target */
1096         rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach);
1097         rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1098
1099         /* Set up fake conntrack:
1100             - to never be deleted, not in any hashes */
1101         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1102         /*  - and look it like as a confirmed connection */
1103         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1104
1105         return ret;
1106
1107 out_fini_proto:
1108         nf_conntrack_proto_fini();
1109 out_free_expect_slab:
1110         kmem_cache_destroy(nf_conntrack_expect_cachep);
1111 err_free_conntrack_slab:
1112         kmem_cache_destroy(nf_conntrack_cachep);
1113 err_free_hash:
1114         free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
1115                             nf_conntrack_htable_size);
1116 err_out:
1117         return -ENOMEM;
1118 }