net/netfilter/nf_conntrack_core.c

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