git.oblomov.eu 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  * 23 Apr 2001: Harald Welte <laforge@gnumonks.org>
  14  *      - new API and handling of conntrack/nat helpers
  15  *      - now capable of multiple expectations for one master
  16  * 16 Jul 2002: Harald Welte <laforge@gnumonks.org>
  17  *      - add usage/reference counts to ip_conntrack_expect
  18  *      - export ip_conntrack[_expect]_{find_get,put} functions
  19  * 16 Dec 2003: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
  20  *      - generalize L3 protocol denendent part.
  21  * 23 Mar 2004: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
  22  *      - add support various size of conntrack structures.
  23  * 26 Jan 2006: Harald Welte <laforge@netfilter.org>
  24  *      - restructure nf_conn (introduce nf_conn_help)
  25  *      - redesign 'features' how they were originally intended
  26  * 26 Feb 2006: Pablo Neira Ayuso <pablo@eurodev.net>
  27  *      - add support for L3 protocol module load on demand.
  28  *
  29  * Derived from net/ipv4/netfilter/ip_conntrack_core.c
  30  */
  31
  32 #include <linux/types.h>
  33 #include <linux/netfilter.h>
  34 #include <linux/module.h>
  35 #include <linux/skbuff.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/vmalloc.h>
  38 #include <linux/stddef.h>
  39 #include <linux/slab.h>
  40 #include <linux/random.h>
  41 #include <linux/jhash.h>
  42 #include <linux/err.h>
  43 #include <linux/percpu.h>
  44 #include <linux/moduleparam.h>
  45 #include <linux/notifier.h>
  46 #include <linux/kernel.h>
  47 #include <linux/netdevice.h>
  48 #include <linux/socket.h>
  49
  50 /* This rwlock protects the main hash table, protocol/helper/expected
  51    registrations, conntrack timers*/
  52 #define ASSERT_READ_LOCK(x)
  53 #define ASSERT_WRITE_LOCK(x)
  54
  55 #include <net/netfilter/nf_conntrack.h>
  56 #include <net/netfilter/nf_conntrack_l3proto.h>
  57 #include <net/netfilter/nf_conntrack_protocol.h>
  58 #include <net/netfilter/nf_conntrack_expect.h>
  59 #include <net/netfilter/nf_conntrack_helper.h>
  60 #include <net/netfilter/nf_conntrack_core.h>
  61
  62 #define NF_CONNTRACK_VERSION    "0.5.0"
  63
  64 #if 0
  65 #define DEBUGP printk
  66 #else
  67 #define DEBUGP(format, args...)
  68 #endif
  69
  70 DEFINE_RWLOCK(nf_conntrack_lock);
  71
  72 /* nf_conntrack_standalone needs this */
  73 atomic_t nf_conntrack_count = ATOMIC_INIT(0);
  74
  75 void (*nf_conntrack_destroyed)(struct nf_conn *conntrack) = NULL;
  76 struct nf_conntrack_protocol **nf_ct_protos[PF_MAX] __read_mostly;
  77 struct nf_conntrack_l3proto *nf_ct_l3protos[PF_MAX] __read_mostly;
  78 static LIST_HEAD(helpers);
  79 unsigned int nf_conntrack_htable_size __read_mostly = 0;
  80 int nf_conntrack_max __read_mostly;
  81 struct list_head *nf_conntrack_hash __read_mostly;
  82 struct nf_conn nf_conntrack_untracked;
  83 unsigned int nf_ct_log_invalid __read_mostly;
  84 static LIST_HEAD(unconfirmed);
  85 static int nf_conntrack_vmalloc __read_mostly;
  86
  87 static unsigned int nf_conntrack_next_id;
  88
  89 #ifdef CONFIG_NF_CONNTRACK_EVENTS
  90 ATOMIC_NOTIFIER_HEAD(nf_conntrack_chain);
  91 ATOMIC_NOTIFIER_HEAD(nf_conntrack_expect_chain);
  92
  93 DEFINE_PER_CPU(struct nf_conntrack_ecache, nf_conntrack_ecache);
  94
  95 /* deliver cached events and clear cache entry - must be called with locally
  96  * disabled softirqs */
  97 static inline void
  98 __nf_ct_deliver_cached_events(struct nf_conntrack_ecache *ecache)
  99 {
 100         DEBUGP("ecache: delivering events for %p\n", ecache->ct);
 101         if (nf_ct_is_confirmed(ecache->ct) && !nf_ct_is_dying(ecache->ct)
 102             && ecache->events)
 103                 atomic_notifier_call_chain(&nf_conntrack_chain, ecache->events,
 104                                     ecache->ct);
 105
 106         ecache->events = 0;
 107         nf_ct_put(ecache->ct);
 108         ecache->ct = NULL;
 109 }
 110
 111 /* Deliver all cached events for a particular conntrack. This is called
 112  * by code prior to async packet handling for freeing the skb */
 113 void nf_ct_deliver_cached_events(const struct nf_conn *ct)
 114 {
 115         struct nf_conntrack_ecache *ecache;
 116
 117         local_bh_disable();
 118         ecache = &__get_cpu_var(nf_conntrack_ecache);
 119         if (ecache->ct == ct)
 120                 __nf_ct_deliver_cached_events(ecache);
 121         local_bh_enable();
 122 }
 123
 124 /* Deliver cached events for old pending events, if current conntrack != old */
 125 void __nf_ct_event_cache_init(struct nf_conn *ct)
 126 {
 127         struct nf_conntrack_ecache *ecache;
 128
 129         /* take care of delivering potentially old events */
 130         ecache = &__get_cpu_var(nf_conntrack_ecache);
 131         BUG_ON(ecache->ct == ct);
 132         if (ecache->ct)
 133                 __nf_ct_deliver_cached_events(ecache);
 134         /* initialize for this conntrack/packet */
 135         ecache->ct = ct;
 136         nf_conntrack_get(&ct->ct_general);
 137 }
 138
 139 /* flush the event cache - touches other CPU's data and must not be called
 140  * while packets are still passing through the code */
 141 static void nf_ct_event_cache_flush(void)
 142 {
 143         struct nf_conntrack_ecache *ecache;
 144         int cpu;
 145
 146         for_each_possible_cpu(cpu) {
 147                 ecache = &per_cpu(nf_conntrack_ecache, cpu);
 148                 if (ecache->ct)
 149                         nf_ct_put(ecache->ct);
 150         }
 151 }
 152 #else
 153 static inline void nf_ct_event_cache_flush(void) {}
 154 #endif /* CONFIG_NF_CONNTRACK_EVENTS */
 155
 156 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
 157 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
 158
 159 /*
 160  * This scheme offers various size of "struct nf_conn" dependent on
 161  * features(helper, nat, ...)
 162  */
 163
 164 #define NF_CT_FEATURES_NAMELEN  256
 165 static struct {
 166         /* name of slab cache. printed in /proc/slabinfo */
 167         char *name;
 168
 169         /* size of slab cache */
 170         size_t size;
 171
 172         /* slab cache pointer */
 173         kmem_cache_t *cachep;
 174
 175         /* allocated slab cache + modules which uses this slab cache */
 176         int use;
 177
 178 } nf_ct_cache[NF_CT_F_NUM];
 179
 180 /* protect members of nf_ct_cache except of "use" */
 181 DEFINE_RWLOCK(nf_ct_cache_lock);
 182
 183 /* This avoids calling kmem_cache_create() with same name simultaneously */
 184 static DEFINE_MUTEX(nf_ct_cache_mutex);
 185
 186 extern struct nf_conntrack_protocol nf_conntrack_generic_protocol;
 187 struct nf_conntrack_protocol *
 188 __nf_ct_proto_find(u_int16_t l3proto, u_int8_t protocol)
 189 {
 190         if (unlikely(l3proto >= AF_MAX || nf_ct_protos[l3proto] == NULL))
 191                 return &nf_conntrack_generic_protocol;
 192
 193         return nf_ct_protos[l3proto][protocol];
 194 }
 195
 196 /* this is guaranteed to always return a valid protocol helper, since
 197  * it falls back to generic_protocol */
 198 struct nf_conntrack_protocol *
 199 nf_ct_proto_find_get(u_int16_t l3proto, u_int8_t protocol)
 200 {
 201         struct nf_conntrack_protocol *p;
 202
 203         preempt_disable();
 204         p = __nf_ct_proto_find(l3proto, protocol);
 205         if (!try_module_get(p->me))
 206                 p = &nf_conntrack_generic_protocol;
 207         preempt_enable();
 208
 209         return p;
 210 }
 211
 212 void nf_ct_proto_put(struct nf_conntrack_protocol *p)
 213 {
 214         module_put(p->me);
 215 }
 216
 217 struct nf_conntrack_l3proto *
 218 nf_ct_l3proto_find_get(u_int16_t l3proto)
 219 {
 220         struct nf_conntrack_l3proto *p;
 221
 222         preempt_disable();
 223         p = __nf_ct_l3proto_find(l3proto);
 224         if (!try_module_get(p->me))
 225                 p = &nf_conntrack_generic_l3proto;
 226         preempt_enable();
 227
 228         return p;
 229 }
 230
 231 void nf_ct_l3proto_put(struct nf_conntrack_l3proto *p)
 232 {
 233         module_put(p->me);
 234 }
 235
 236 int
 237 nf_ct_l3proto_try_module_get(unsigned short l3proto)
 238 {
 239         int ret;
 240         struct nf_conntrack_l3proto *p;
 241
 242 retry:  p = nf_ct_l3proto_find_get(l3proto);
 243         if (p == &nf_conntrack_generic_l3proto) {
 244                 ret = request_module("nf_conntrack-%d", l3proto);
 245                 if (!ret)
 246                         goto retry;
 247
 248                 return -EPROTOTYPE;
 249         }
 250
 251         return 0;
 252 }
 253
 254 void nf_ct_l3proto_module_put(unsigned short l3proto)
 255 {
 256         struct nf_conntrack_l3proto *p;
 257
 258         preempt_disable();
 259         p = __nf_ct_l3proto_find(l3proto);
 260         preempt_enable();
 261
 262         module_put(p->me);
 263 }
 264
 265 static int nf_conntrack_hash_rnd_initted;
 266 static unsigned int nf_conntrack_hash_rnd;
 267
 268 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
 269                                   unsigned int size, unsigned int rnd)
 270 {
 271         unsigned int a, b;
 272         a = jhash((void *)tuple->src.u3.all, sizeof(tuple->src.u3.all),
 273                   ((tuple->src.l3num) << 16) | tuple->dst.protonum);
 274         b = jhash((void *)tuple->dst.u3.all, sizeof(tuple->dst.u3.all),
 275                         (tuple->src.u.all << 16) | tuple->dst.u.all);
 276
 277         return jhash_2words(a, b, rnd) % size;
 278 }
 279
 280 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
 281 {
 282         return __hash_conntrack(tuple, nf_conntrack_htable_size,
 283                                 nf_conntrack_hash_rnd);
 284 }
 285
 286 int nf_conntrack_register_cache(u_int32_t features, const char *name,
 287                                 size_t size)
 288 {
 289         int ret = 0;
 290         char *cache_name;
 291         kmem_cache_t *cachep;
 292
 293         DEBUGP("nf_conntrack_register_cache: features=0x%x, name=%s, size=%d\n",
 294                features, name, size);
 295
 296         if (features < NF_CT_F_BASIC || features >= NF_CT_F_NUM) {
 297                 DEBUGP("nf_conntrack_register_cache: invalid features.: 0x%x\n",
 298                         features);
 299                 return -EINVAL;
 300         }
 301
 302         mutex_lock(&nf_ct_cache_mutex);
 303
 304         write_lock_bh(&nf_ct_cache_lock);
 305         /* e.g: multiple helpers are loaded */
 306         if (nf_ct_cache[features].use > 0) {
 307                 DEBUGP("nf_conntrack_register_cache: already resisterd.\n");
 308                 if ((!strncmp(nf_ct_cache[features].name, name,
 309                               NF_CT_FEATURES_NAMELEN))
 310                     && nf_ct_cache[features].size == size) {
 311                         DEBUGP("nf_conntrack_register_cache: reusing.\n");
 312                         nf_ct_cache[features].use++;
 313                         ret = 0;
 314                 } else
 315                         ret = -EBUSY;
 316
 317                 write_unlock_bh(&nf_ct_cache_lock);
 318                 mutex_unlock(&nf_ct_cache_mutex);
 319                 return ret;
 320         }
 321         write_unlock_bh(&nf_ct_cache_lock);
 322
 323         /*
 324          * The memory space for name of slab cache must be alive until
 325          * cache is destroyed.
 326          */
 327         cache_name = kmalloc(sizeof(char)*NF_CT_FEATURES_NAMELEN, GFP_ATOMIC);
 328         if (cache_name == NULL) {
 329                 DEBUGP("nf_conntrack_register_cache: can't alloc cache_name\n");
 330                 ret = -ENOMEM;
 331                 goto out_up_mutex;
 332         }
 333
 334         if (strlcpy(cache_name, name, NF_CT_FEATURES_NAMELEN)
 335                                                 >= NF_CT_FEATURES_NAMELEN) {
 336                 printk("nf_conntrack_register_cache: name too long\n");
 337                 ret = -EINVAL;
 338                 goto out_free_name;
 339         }
 340
 341         cachep = kmem_cache_create(cache_name, size, 0, 0,
 342                                    NULL, NULL);
 343         if (!cachep) {
 344                 printk("nf_conntrack_register_cache: Can't create slab cache "
 345                        "for the features = 0x%x\n", features);
 346                 ret = -ENOMEM;
 347                 goto out_free_name;
 348         }
 349
 350         write_lock_bh(&nf_ct_cache_lock);
 351         nf_ct_cache[features].use = 1;
 352         nf_ct_cache[features].size = size;
 353         nf_ct_cache[features].cachep = cachep;
 354         nf_ct_cache[features].name = cache_name;
 355         write_unlock_bh(&nf_ct_cache_lock);
 356
 357         goto out_up_mutex;
 358
 359 out_free_name:
 360         kfree(cache_name);
 361 out_up_mutex:
 362         mutex_unlock(&nf_ct_cache_mutex);
 363         return ret;
 364 }
 365
 366 /* FIXME: In the current, only nf_conntrack_cleanup() can call this function. */
 367 void nf_conntrack_unregister_cache(u_int32_t features)
 368 {
 369         kmem_cache_t *cachep;
 370         char *name;
 371
 372         /*
 373          * This assures that kmem_cache_create() isn't called before destroying
 374          * slab cache.
 375          */
 376         DEBUGP("nf_conntrack_unregister_cache: 0x%04x\n", features);
 377         mutex_lock(&nf_ct_cache_mutex);
 378
 379         write_lock_bh(&nf_ct_cache_lock);
 380         if (--nf_ct_cache[features].use > 0) {
 381                 write_unlock_bh(&nf_ct_cache_lock);
 382                 mutex_unlock(&nf_ct_cache_mutex);
 383                 return;
 384         }
 385         cachep = nf_ct_cache[features].cachep;
 386         name = nf_ct_cache[features].name;
 387         nf_ct_cache[features].cachep = NULL;
 388         nf_ct_cache[features].name = NULL;
 389         nf_ct_cache[features].size = 0;
 390         write_unlock_bh(&nf_ct_cache_lock);
 391
 392         synchronize_net();
 393
 394         kmem_cache_destroy(cachep);
 395         kfree(name);
 396
 397         mutex_unlock(&nf_ct_cache_mutex);
 398 }
 399
 400 int
 401 nf_ct_get_tuple(const struct sk_buff *skb,
 402                 unsigned int nhoff,
 403                 unsigned int dataoff,
 404                 u_int16_t l3num,
 405                 u_int8_t protonum,
 406                 struct nf_conntrack_tuple *tuple,
 407                 const struct nf_conntrack_l3proto *l3proto,
 408                 const struct nf_conntrack_protocol *protocol)
 409 {
 410         NF_CT_TUPLE_U_BLANK(tuple);
 411
 412         tuple->src.l3num = l3num;
 413         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 414                 return 0;
 415
 416         tuple->dst.protonum = protonum;
 417         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 418
 419         return protocol->pkt_to_tuple(skb, dataoff, tuple);
 420 }
 421
 422 int
 423 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 424                    const struct nf_conntrack_tuple *orig,
 425                    const struct nf_conntrack_l3proto *l3proto,
 426                    const struct nf_conntrack_protocol *protocol)
 427 {
 428         NF_CT_TUPLE_U_BLANK(inverse);
 429
 430         inverse->src.l3num = orig->src.l3num;
 431         if (l3proto->invert_tuple(inverse, orig) == 0)
 432                 return 0;
 433
 434         inverse->dst.dir = !orig->dst.dir;
 435
 436         inverse->dst.protonum = orig->dst.protonum;
 437         return protocol->invert_tuple(inverse, orig);
 438 }
 439
 440 static void
 441 clean_from_lists(struct nf_conn *ct)
 442 {
 443         DEBUGP("clean_from_lists(%p)\n", ct);
 444         ASSERT_WRITE_LOCK(&nf_conntrack_lock);
 445         list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
 446         list_del(&ct->tuplehash[IP_CT_DIR_REPLY].list);
 447
 448         /* Destroy all pending expectations */
 449         nf_ct_remove_expectations(ct);
 450 }
 451
 452 static void
 453 destroy_conntrack(struct nf_conntrack *nfct)
 454 {
 455         struct nf_conn *ct = (struct nf_conn *)nfct;
 456         struct nf_conntrack_l3proto *l3proto;
 457         struct nf_conntrack_protocol *proto;
 458
 459         DEBUGP("destroy_conntrack(%p)\n", ct);
 460         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 461         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 462
 463         nf_conntrack_event(IPCT_DESTROY, ct);
 464         set_bit(IPS_DYING_BIT, &ct->status);
 465
 466         /* To make sure we don't get any weird locking issues here:
 467          * destroy_conntrack() MUST NOT be called with a write lock
 468          * to nf_conntrack_lock!!! -HW */
 469         l3proto = __nf_ct_l3proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num);
 470         if (l3proto && l3proto->destroy)
 471                 l3proto->destroy(ct);
 472
 473         proto = __nf_ct_proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num, ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
 474         if (proto && proto->destroy)
 475                 proto->destroy(ct);
 476
 477         if (nf_conntrack_destroyed)
 478                 nf_conntrack_destroyed(ct);
 479
 480         write_lock_bh(&nf_conntrack_lock);
 481         /* Expectations will have been removed in clean_from_lists,
 482          * except TFTP can create an expectation on the first packet,
 483          * before connection is in the list, so we need to clean here,
 484          * too. */
 485         nf_ct_remove_expectations(ct);
 486
 487         /* We overload first tuple to link into unconfirmed list. */
 488         if (!nf_ct_is_confirmed(ct)) {
 489                 BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list));
 490                 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
 491         }
 492
 493         NF_CT_STAT_INC(delete);
 494         write_unlock_bh(&nf_conntrack_lock);
 495
 496         if (ct->master)
 497                 nf_ct_put(ct->master);
 498
 499         DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct);
 500         nf_conntrack_free(ct);
 501 }
 502
 503 static void death_by_timeout(unsigned long ul_conntrack)
 504 {
 505         struct nf_conn *ct = (void *)ul_conntrack;
 506
 507         write_lock_bh(&nf_conntrack_lock);
 508         /* Inside lock so preempt is disabled on module removal path.
 509          * Otherwise we can get spurious warnings. */
 510         NF_CT_STAT_INC(delete_list);
 511         clean_from_lists(ct);
 512         write_unlock_bh(&nf_conntrack_lock);
 513         nf_ct_put(ct);
 514 }
 515
 516 struct nf_conntrack_tuple_hash *
 517 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple,
 518                     const struct nf_conn *ignored_conntrack)
 519 {
 520         struct nf_conntrack_tuple_hash *h;
 521         unsigned int hash = hash_conntrack(tuple);
 522
 523         ASSERT_READ_LOCK(&nf_conntrack_lock);
 524         list_for_each_entry(h, &nf_conntrack_hash[hash], list) {
 525                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
 526                     nf_ct_tuple_equal(tuple, &h->tuple)) {
 527                         NF_CT_STAT_INC(found);
 528                         return h;
 529                 }
 530                 NF_CT_STAT_INC(searched);
 531         }
 532
 533         return NULL;
 534 }
 535
 536 /* Find a connection corresponding to a tuple. */
 537 struct nf_conntrack_tuple_hash *
 538 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple,
 539                       const struct nf_conn *ignored_conntrack)
 540 {
 541         struct nf_conntrack_tuple_hash *h;
 542
 543         read_lock_bh(&nf_conntrack_lock);
 544         h = __nf_conntrack_find(tuple, ignored_conntrack);
 545         if (h)
 546                 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
 547         read_unlock_bh(&nf_conntrack_lock);
 548
 549         return h;
 550 }
 551
 552 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 553                                        unsigned int hash,
 554                                        unsigned int repl_hash)
 555 {
 556         ct->id = ++nf_conntrack_next_id;
 557         list_add(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list,
 558                  &nf_conntrack_hash[hash]);
 559         list_add(&ct->tuplehash[IP_CT_DIR_REPLY].list,
 560                  &nf_conntrack_hash[repl_hash]);
 561 }
 562
 563 void nf_conntrack_hash_insert(struct nf_conn *ct)
 564 {
 565         unsigned int hash, repl_hash;
 566
 567         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 568         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 569
 570         write_lock_bh(&nf_conntrack_lock);
 571         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 572         write_unlock_bh(&nf_conntrack_lock);
 573 }
 574
 575 /* Confirm a connection given skb; places it in hash table */
 576 int
 577 __nf_conntrack_confirm(struct sk_buff **pskb)
 578 {
 579         unsigned int hash, repl_hash;
 580         struct nf_conntrack_tuple_hash *h;
 581         struct nf_conn *ct;
 582         struct nf_conn_help *help;
 583         enum ip_conntrack_info ctinfo;
 584
 585         ct = nf_ct_get(*pskb, &ctinfo);
 586
 587         /* ipt_REJECT uses nf_conntrack_attach to attach related
 588            ICMP/TCP RST packets in other direction.  Actual packet
 589            which created connection will be IP_CT_NEW or for an
 590            expected connection, IP_CT_RELATED. */
 591         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 592                 return NF_ACCEPT;
 593
 594         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 595         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 596
 597         /* We're not in hash table, and we refuse to set up related
 598            connections for unconfirmed conns.  But packet copies and
 599            REJECT will give spurious warnings here. */
 600         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 601
 602         /* No external references means noone else could have
 603            confirmed us. */
 604         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 605         DEBUGP("Confirming conntrack %p\n", ct);
 606
 607         write_lock_bh(&nf_conntrack_lock);
 608
 609         /* See if there's one in the list already, including reverse:
 610            NAT could have grabbed it without realizing, since we're
 611            not in the hash.  If there is, we lost race. */
 612         list_for_each_entry(h, &nf_conntrack_hash[hash], list)
 613                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 614                                       &h->tuple))
 615                         goto out;
 616         list_for_each_entry(h, &nf_conntrack_hash[repl_hash], list)
 617                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 618                                       &h->tuple))
 619                         goto out;
 620
 621         /* Remove from unconfirmed list */
 622         list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
 623
 624         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 625         /* Timer relative to confirmation time, not original
 626            setting time, otherwise we'd get timer wrap in
 627            weird delay cases. */
 628         ct->timeout.expires += jiffies;
 629         add_timer(&ct->timeout);
 630         atomic_inc(&ct->ct_general.use);
 631         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 632         NF_CT_STAT_INC(insert);
 633         write_unlock_bh(&nf_conntrack_lock);
 634         help = nfct_help(ct);
 635         if (help && help->helper)
 636                 nf_conntrack_event_cache(IPCT_HELPER, *pskb);
 637 #ifdef CONFIG_NF_NAT_NEEDED
 638         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
 639             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
 640                 nf_conntrack_event_cache(IPCT_NATINFO, *pskb);
 641 #endif
 642         nf_conntrack_event_cache(master_ct(ct) ?
 643                                  IPCT_RELATED : IPCT_NEW, *pskb);
 644         return NF_ACCEPT;
 645
 646 out:
 647         NF_CT_STAT_INC(insert_failed);
 648         write_unlock_bh(&nf_conntrack_lock);
 649         return NF_DROP;
 650 }
 651
 652 /* Returns true if a connection correspondings to the tuple (required
 653    for NAT). */
 654 int
 655 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 656                          const struct nf_conn *ignored_conntrack)
 657 {
 658         struct nf_conntrack_tuple_hash *h;
 659
 660         read_lock_bh(&nf_conntrack_lock);
 661         h = __nf_conntrack_find(tuple, ignored_conntrack);
 662         read_unlock_bh(&nf_conntrack_lock);
 663
 664         return h != NULL;
 665 }
 666
 667 /* There's a small race here where we may free a just-assured
 668    connection.  Too bad: we're in trouble anyway. */
 669 static int early_drop(struct list_head *chain)
 670 {
 671         /* Traverse backwards: gives us oldest, which is roughly LRU */
 672         struct nf_conntrack_tuple_hash *h;
 673         struct nf_conn *ct = NULL, *tmp;
 674         int dropped = 0;
 675
 676         read_lock_bh(&nf_conntrack_lock);
 677         list_for_each_entry_reverse(h, chain, list) {
 678                 tmp = nf_ct_tuplehash_to_ctrack(h);
 679                 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) {
 680                         ct = tmp;
 681                         atomic_inc(&ct->ct_general.use);
 682                         break;
 683                 }
 684         }
 685         read_unlock_bh(&nf_conntrack_lock);
 686
 687         if (!ct)
 688                 return dropped;
 689
 690         if (del_timer(&ct->timeout)) {
 691                 death_by_timeout((unsigned long)ct);
 692                 dropped = 1;
 693                 NF_CT_STAT_INC(early_drop);
 694         }
 695         nf_ct_put(ct);
 696         return dropped;
 697 }
 698
 699 static struct nf_conntrack_helper *
 700 __nf_ct_helper_find(const struct nf_conntrack_tuple *tuple)
 701 {
 702         struct nf_conntrack_helper *h;
 703
 704         list_for_each_entry(h, &helpers, list) {
 705                 if (nf_ct_tuple_mask_cmp(tuple, &h->tuple, &h->mask))
 706                         return h;
 707         }
 708         return NULL;
 709 }
 710
 711 struct nf_conntrack_helper *
 712 nf_ct_helper_find_get( const struct nf_conntrack_tuple *tuple)
 713 {
 714         struct nf_conntrack_helper *helper;
 715
 716         /* need nf_conntrack_lock to assure that helper exists until
 717          * try_module_get() is called */
 718         read_lock_bh(&nf_conntrack_lock);
 719
 720         helper = __nf_ct_helper_find(tuple);
 721         if (helper) {
 722                 /* need to increase module usage count to assure helper will
 723                  * not go away while the caller is e.g. busy putting a
 724                  * conntrack in the hash that uses the helper */
 725                 if (!try_module_get(helper->me))
 726                         helper = NULL;
 727         }
 728
 729         read_unlock_bh(&nf_conntrack_lock);
 730
 731         return helper;
 732 }
 733
 734 void nf_ct_helper_put(struct nf_conntrack_helper *helper)
 735 {
 736         module_put(helper->me);
 737 }
 738
 739 static struct nf_conn *
 740 __nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
 741                      const struct nf_conntrack_tuple *repl,
 742                      const struct nf_conntrack_l3proto *l3proto)
 743 {
 744         struct nf_conn *conntrack = NULL;
 745         u_int32_t features = 0;
 746         struct nf_conntrack_helper *helper;
 747
 748         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 749                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
 750                 nf_conntrack_hash_rnd_initted = 1;
 751         }
 752
 753         /* We don't want any race condition at early drop stage */
 754         atomic_inc(&nf_conntrack_count);
 755
 756         if (nf_conntrack_max
 757             && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
 758                 unsigned int hash = hash_conntrack(orig);
 759                 /* Try dropping from this hash chain. */
 760                 if (!early_drop(&nf_conntrack_hash[hash])) {
 761                         atomic_dec(&nf_conntrack_count);
 762                         if (net_ratelimit())
 763                                 printk(KERN_WARNING
 764                                        "nf_conntrack: table full, dropping"
 765                                        " packet.\n");
 766                         return ERR_PTR(-ENOMEM);
 767                 }
 768         }
 769
 770         /*  find features needed by this conntrack. */
 771         features = l3proto->get_features(orig);
 772
 773         /* FIXME: protect helper list per RCU */
 774         read_lock_bh(&nf_conntrack_lock);
 775         helper = __nf_ct_helper_find(repl);
 776         if (helper)
 777                 features |= NF_CT_F_HELP;
 778         read_unlock_bh(&nf_conntrack_lock);
 779
 780         DEBUGP("nf_conntrack_alloc: features=0x%x\n", features);
 781
 782         read_lock_bh(&nf_ct_cache_lock);
 783
 784         if (unlikely(!nf_ct_cache[features].use)) {
 785                 DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n",
 786                         features);
 787                 goto out;
 788         }
 789
 790         conntrack = kmem_cache_alloc(nf_ct_cache[features].cachep, GFP_ATOMIC);
 791         if (conntrack == NULL) {
 792                 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n");
 793                 goto out;
 794         }
 795
 796         memset(conntrack, 0, nf_ct_cache[features].size);
 797         conntrack->features = features;
 798         atomic_set(&conntrack->ct_general.use, 1);
 799         conntrack->ct_general.destroy = destroy_conntrack;
 800         conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 801         conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 802         /* Don't set timer yet: wait for confirmation */
 803         init_timer(&conntrack->timeout);
 804         conntrack->timeout.data = (unsigned long)conntrack;
 805         conntrack->timeout.function = death_by_timeout;
 806         read_unlock_bh(&nf_ct_cache_lock);
 807
 808         return conntrack;
 809 out:
 810         read_unlock_bh(&nf_ct_cache_lock);
 811         atomic_dec(&nf_conntrack_count);
 812         return conntrack;
 813 }
 814
 815 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
 816                                    const struct nf_conntrack_tuple *repl)
 817 {
 818         struct nf_conntrack_l3proto *l3proto;
 819
 820         l3proto = __nf_ct_l3proto_find(orig->src.l3num);
 821         return __nf_conntrack_alloc(orig, repl, l3proto);
 822 }
 823
 824 void nf_conntrack_free(struct nf_conn *conntrack)
 825 {
 826         u_int32_t features = conntrack->features;
 827         NF_CT_ASSERT(features >= NF_CT_F_BASIC && features < NF_CT_F_NUM);
 828         DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features,
 829                conntrack);
 830         kmem_cache_free(nf_ct_cache[features].cachep, conntrack);
 831         atomic_dec(&nf_conntrack_count);
 832 }
 833
 834 /* Allocate a new conntrack: we return -ENOMEM if classification
 835    failed due to stress.  Otherwise it really is unclassifiable. */
 836 static struct nf_conntrack_tuple_hash *
 837 init_conntrack(const struct nf_conntrack_tuple *tuple,
 838                struct nf_conntrack_l3proto *l3proto,
 839                struct nf_conntrack_protocol *protocol,
 840                struct sk_buff *skb,
 841                unsigned int dataoff)
 842 {
 843         struct nf_conn *conntrack;
 844         struct nf_conntrack_tuple repl_tuple;
 845         struct nf_conntrack_expect *exp;
 846
 847         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, protocol)) {
 848                 DEBUGP("Can't invert tuple.\n");
 849                 return NULL;
 850         }
 851
 852         conntrack = __nf_conntrack_alloc(tuple, &repl_tuple, l3proto);
 853         if (conntrack == NULL || IS_ERR(conntrack)) {
 854                 DEBUGP("Can't allocate conntrack.\n");
 855                 return (struct nf_conntrack_tuple_hash *)conntrack;
 856         }
 857
 858         if (!protocol->new(conntrack, skb, dataoff)) {
 859                 nf_conntrack_free(conntrack);
 860                 DEBUGP("init conntrack: can't track with proto module\n");
 861                 return NULL;
 862         }
 863
 864         write_lock_bh(&nf_conntrack_lock);
 865         exp = find_expectation(tuple);
 866
 867         if (exp) {
 868                 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
 869                         conntrack, exp);
 870                 /* Welcome, Mr. Bond.  We've been expecting you... */
 871                 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
 872                 conntrack->master = exp->master;
 873 #ifdef CONFIG_NF_CONNTRACK_MARK
 874                 conntrack->mark = exp->master->mark;
 875 #endif
 876 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 877                 conntrack->secmark = exp->master->secmark;
 878 #endif
 879                 nf_conntrack_get(&conntrack->master->ct_general);
 880                 NF_CT_STAT_INC(expect_new);
 881         } else {
 882                 struct nf_conn_help *help = nfct_help(conntrack);
 883
 884                 if (help)
 885                         help->helper = __nf_ct_helper_find(&repl_tuple);
 886                 NF_CT_STAT_INC(new);
 887         }
 888
 889         /* Overload tuple linked list to put us in unconfirmed list. */
 890         list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed);
 891
 892         write_unlock_bh(&nf_conntrack_lock);
 893
 894         if (exp) {
 895                 if (exp->expectfn)
 896                         exp->expectfn(conntrack, exp);
 897                 nf_conntrack_expect_put(exp);
 898         }
 899
 900         return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
 901 }
 902
 903 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 904 static inline struct nf_conn *
 905 resolve_normal_ct(struct sk_buff *skb,
 906                   unsigned int dataoff,
 907                   u_int16_t l3num,
 908                   u_int8_t protonum,
 909                   struct nf_conntrack_l3proto *l3proto,
 910                   struct nf_conntrack_protocol *proto,
 911                   int *set_reply,
 912                   enum ip_conntrack_info *ctinfo)
 913 {
 914         struct nf_conntrack_tuple tuple;
 915         struct nf_conntrack_tuple_hash *h;
 916         struct nf_conn *ct;
 917
 918         if (!nf_ct_get_tuple(skb, (unsigned int)(skb->nh.raw - skb->data),
 919                              dataoff, l3num, protonum, &tuple, l3proto,
 920                              proto)) {
 921                 DEBUGP("resolve_normal_ct: Can't get tuple\n");
 922                 return NULL;
 923         }
 924
 925         /* look for tuple match */
 926         h = nf_conntrack_find_get(&tuple, NULL);
 927         if (!h) {
 928                 h = init_conntrack(&tuple, l3proto, proto, skb, dataoff);
 929                 if (!h)
 930                         return NULL;
 931                 if (IS_ERR(h))
 932                         return (void *)h;
 933         }
 934         ct = nf_ct_tuplehash_to_ctrack(h);
 935
 936         /* It exists; we have (non-exclusive) reference. */
 937         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 938                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 939                 /* Please set reply bit if this packet OK */
 940                 *set_reply = 1;
 941         } else {
 942                 /* Once we've had two way comms, always ESTABLISHED. */
 943                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 944                         DEBUGP("nf_conntrack_in: normal packet for %p\n", ct);
 945                         *ctinfo = IP_CT_ESTABLISHED;
 946                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 947                         DEBUGP("nf_conntrack_in: related packet for %p\n", ct);
 948                         *ctinfo = IP_CT_RELATED;
 949                 } else {
 950                         DEBUGP("nf_conntrack_in: new packet for %p\n", ct);
 951                         *ctinfo = IP_CT_NEW;
 952                 }
 953                 *set_reply = 0;
 954         }
 955         skb->nfct = &ct->ct_general;
 956         skb->nfctinfo = *ctinfo;
 957         return ct;
 958 }
 959
 960 unsigned int
 961 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb)
 962 {
 963         struct nf_conn *ct;
 964         enum ip_conntrack_info ctinfo;
 965         struct nf_conntrack_l3proto *l3proto;
 966         struct nf_conntrack_protocol *proto;
 967         unsigned int dataoff;
 968         u_int8_t protonum;
 969         int set_reply = 0;
 970         int ret;
 971
 972         /* Previously seen (loopback or untracked)?  Ignore. */
 973         if ((*pskb)->nfct) {
 974                 NF_CT_STAT_INC(ignore);
 975                 return NF_ACCEPT;
 976         }
 977
 978         l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
 979         if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) {
 980                 DEBUGP("not prepared to track yet or error occured\n");
 981                 return -ret;
 982         }
 983
 984         proto = __nf_ct_proto_find((u_int16_t)pf, protonum);
 985
 986         /* It may be an special packet, error, unclean...
 987          * inverse of the return code tells to the netfilter
 988          * core what to do with the packet. */
 989         if (proto->error != NULL &&
 990             (ret = proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
 991                 NF_CT_STAT_INC(error);
 992                 NF_CT_STAT_INC(invalid);
 993                 return -ret;
 994         }
 995
 996         ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, proto,
 997                                &set_reply, &ctinfo);
 998         if (!ct) {
 999                 /* Not valid part of a connection */
1000                 NF_CT_STAT_INC(invalid);
1001                 return NF_ACCEPT;
1002         }
1003
1004         if (IS_ERR(ct)) {
1005                 /* Too stressed to deal. */
1006                 NF_CT_STAT_INC(drop);
1007                 return NF_DROP;
1008         }
1009
1010         NF_CT_ASSERT((*pskb)->nfct);
1011
1012         ret = proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum);
1013         if (ret < 0) {
1014                 /* Invalid: inverse of the return code tells
1015                  * the netfilter core what to do */
1016                 DEBUGP("nf_conntrack_in: Can't track with proto module\n");
1017                 nf_conntrack_put((*pskb)->nfct);
1018                 (*pskb)->nfct = NULL;
1019                 NF_CT_STAT_INC(invalid);
1020                 return -ret;
1021         }
1022
1023         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1024                 nf_conntrack_event_cache(IPCT_STATUS, *pskb);
1025
1026         return ret;
1027 }
1028
1029 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1030                          const struct nf_conntrack_tuple *orig)
1031 {
1032         return nf_ct_invert_tuple(inverse, orig,
1033                                   __nf_ct_l3proto_find(orig->src.l3num),
1034                                   __nf_ct_proto_find(orig->src.l3num,
1035                                                      orig->dst.protonum));
1036 }
1037
1038 int nf_conntrack_helper_register(struct nf_conntrack_helper *me)
1039 {
1040         int ret;
1041         BUG_ON(me->timeout == 0);
1042
1043         ret = nf_conntrack_register_cache(NF_CT_F_HELP, "nf_conntrack:help",
1044                                           sizeof(struct nf_conn)
1045                                           + sizeof(struct nf_conn_help)
1046                                           + __alignof__(struct nf_conn_help));
1047         if (ret < 0) {
1048                 printk(KERN_ERR "nf_conntrack_helper_reigster: Unable to create slab cache for conntracks\n");
1049                 return ret;
1050         }
1051         write_lock_bh(&nf_conntrack_lock);
1052         list_add(&me->list, &helpers);
1053         write_unlock_bh(&nf_conntrack_lock);
1054
1055         return 0;
1056 }
1057
1058 struct nf_conntrack_helper *
1059 __nf_conntrack_helper_find_byname(const char *name)
1060 {
1061         struct nf_conntrack_helper *h;
1062
1063         list_for_each_entry(h, &helpers, list) {
1064                 if (!strcmp(h->name, name))
1065                         return h;
1066         }
1067
1068         return NULL;
1069 }
1070
1071 static inline void unhelp(struct nf_conntrack_tuple_hash *i,
1072                           const struct nf_conntrack_helper *me)
1073 {
1074         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(i);
1075         struct nf_conn_help *help = nfct_help(ct);
1076
1077         if (help && help->helper == me) {
1078                 nf_conntrack_event(IPCT_HELPER, ct);
1079                 help->helper = NULL;
1080         }
1081 }
1082
1083 void nf_conntrack_helper_unregister(struct nf_conntrack_helper *me)
1084 {
1085         unsigned int i;
1086         struct nf_conntrack_tuple_hash *h;
1087         struct nf_conntrack_expect *exp, *tmp;
1088
1089         /* Need write lock here, to delete helper. */
1090         write_lock_bh(&nf_conntrack_lock);
1091         list_del(&me->list);
1092
1093         /* Get rid of expectations */
1094         list_for_each_entry_safe(exp, tmp, &nf_conntrack_expect_list, list) {
1095                 struct nf_conn_help *help = nfct_help(exp->master);
1096                 if (help->helper == me && del_timer(&exp->timeout)) {
1097                         nf_ct_unlink_expect(exp);
1098                         nf_conntrack_expect_put(exp);
1099                 }
1100         }
1101
1102         /* Get rid of expecteds, set helpers to NULL. */
1103         list_for_each_entry(h, &unconfirmed, list)
1104                 unhelp(h, me);
1105         for (i = 0; i < nf_conntrack_htable_size; i++) {
1106                 list_for_each_entry(h, &nf_conntrack_hash[i], list)
1107                         unhelp(h, me);
1108         }
1109         write_unlock_bh(&nf_conntrack_lock);
1110
1111         /* Someone could be still looking at the helper in a bh. */
1112         synchronize_net();
1113 }
1114
1115 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1116 void __nf_ct_refresh_acct(struct nf_conn *ct,
1117                           enum ip_conntrack_info ctinfo,
1118                           const struct sk_buff *skb,
1119                           unsigned long extra_jiffies,
1120                           int do_acct)
1121 {
1122         int event = 0;
1123
1124         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1125         NF_CT_ASSERT(skb);
1126
1127         write_lock_bh(&nf_conntrack_lock);
1128
1129         /* Only update if this is not a fixed timeout */
1130         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) {
1131                 write_unlock_bh(&nf_conntrack_lock);
1132                 return;
1133         }
1134
1135         /* If not in hash table, timer will not be active yet */
1136         if (!nf_ct_is_confirmed(ct)) {
1137                 ct->timeout.expires = extra_jiffies;
1138                 event = IPCT_REFRESH;
1139         } else {
1140                 /* Need del_timer for race avoidance (may already be dying). */
1141                 if (del_timer(&ct->timeout)) {
1142                         ct->timeout.expires = jiffies + extra_jiffies;
1143                         add_timer(&ct->timeout);
1144                         event = IPCT_REFRESH;
1145                 }
1146         }
1147
1148 #ifdef CONFIG_NF_CT_ACCT
1149         if (do_acct) {
1150                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
1151                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
1152                         skb->len - (unsigned int)(skb->nh.raw - skb->data);
1153         if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
1154             || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
1155                 event |= IPCT_COUNTER_FILLING;
1156         }
1157 #endif
1158
1159         write_unlock_bh(&nf_conntrack_lock);
1160
1161         /* must be unlocked when calling event cache */
1162         if (event)
1163                 nf_conntrack_event_cache(event, skb);
1164 }
1165
1166 #if defined(CONFIG_NF_CT_NETLINK) || \
1167     defined(CONFIG_NF_CT_NETLINK_MODULE)
1168
1169 #include <linux/netfilter/nfnetlink.h>
1170 #include <linux/netfilter/nfnetlink_conntrack.h>
1171 #include <linux/mutex.h>
1172
1173
1174 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1175  * in ip_conntrack_core, since we don't want the protocols to autoload
1176  * or depend on ctnetlink */
1177 int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb,
1178                                const struct nf_conntrack_tuple *tuple)
1179 {
1180         NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t),
1181                 &tuple->src.u.tcp.port);
1182         NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t),
1183                 &tuple->dst.u.tcp.port);
1184         return 0;
1185
1186 nfattr_failure:
1187         return -1;
1188 }
1189
1190 static const size_t cta_min_proto[CTA_PROTO_MAX] = {
1191         [CTA_PROTO_SRC_PORT-1]  = sizeof(u_int16_t),
1192         [CTA_PROTO_DST_PORT-1]  = sizeof(u_int16_t)
1193 };
1194
1195 int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[],
1196                                struct nf_conntrack_tuple *t)
1197 {
1198         if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1])
1199                 return -EINVAL;
1200
1201         if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
1202                 return -EINVAL;
1203
1204         t->src.u.tcp.port =
1205                 *(u_int16_t *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]);
1206         t->dst.u.tcp.port =
1207                 *(u_int16_t *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]);
1208
1209         return 0;
1210 }
1211 #endif
1212
1213 /* Used by ipt_REJECT and ip6t_REJECT. */
1214 void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1215 {
1216         struct nf_conn *ct;
1217         enum ip_conntrack_info ctinfo;
1218
1219         /* This ICMP is in reverse direction to the packet which caused it */
1220         ct = nf_ct_get(skb, &ctinfo);
1221         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1222                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
1223         else
1224                 ctinfo = IP_CT_RELATED;
1225
1226         /* Attach to new skbuff, and increment count */
1227         nskb->nfct = &ct->ct_general;
1228         nskb->nfctinfo = ctinfo;
1229         nf_conntrack_get(nskb->nfct);
1230 }
1231
1232 static inline int
1233 do_iter(const struct nf_conntrack_tuple_hash *i,
1234         int (*iter)(struct nf_conn *i, void *data),
1235         void *data)
1236 {
1237         return iter(nf_ct_tuplehash_to_ctrack(i), data);
1238 }
1239
1240 /* Bring out ya dead! */
1241 static struct nf_conn *
1242 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
1243                 void *data, unsigned int *bucket)
1244 {
1245         struct nf_conntrack_tuple_hash *h;
1246         struct nf_conn *ct;
1247
1248         write_lock_bh(&nf_conntrack_lock);
1249         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1250                 list_for_each_entry(h, &nf_conntrack_hash[*bucket], list) {
1251                         ct = nf_ct_tuplehash_to_ctrack(h);
1252                         if (iter(ct, data))
1253                                 goto found;
1254                 }
1255         }
1256         list_for_each_entry(h, &unconfirmed, list) {
1257                 ct = nf_ct_tuplehash_to_ctrack(h);
1258                 if (iter(ct, data))
1259                         goto found;
1260         }
1261         write_unlock_bh(&nf_conntrack_lock);
1262         return NULL;
1263 found:
1264         atomic_inc(&ct->ct_general.use);
1265         write_unlock_bh(&nf_conntrack_lock);
1266         return ct;
1267 }
1268
1269 void
1270 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
1271 {
1272         struct nf_conn *ct;
1273         unsigned int bucket = 0;
1274
1275         while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
1276                 /* Time to push up daises... */
1277                 if (del_timer(&ct->timeout))
1278                         death_by_timeout((unsigned long)ct);
1279                 /* ... else the timer will get him soon. */
1280
1281                 nf_ct_put(ct);
1282         }
1283 }
1284
1285 static int kill_all(struct nf_conn *i, void *data)
1286 {
1287         return 1;
1288 }
1289
1290 static void free_conntrack_hash(struct list_head *hash, int vmalloced, int size)
1291 {
1292         if (vmalloced)
1293                 vfree(hash);
1294         else
1295                 free_pages((unsigned long)hash,
1296                            get_order(sizeof(struct list_head) * size));
1297 }
1298
1299 void nf_conntrack_flush()
1300 {
1301         nf_ct_iterate_cleanup(kill_all, NULL);
1302 }
1303
1304 /* Mishearing the voices in his head, our hero wonders how he's
1305    supposed to kill the mall. */
1306 void nf_conntrack_cleanup(void)
1307 {
1308         int i;
1309
1310         ip_ct_attach = NULL;
1311
1312         /* This makes sure all current packets have passed through
1313            netfilter framework.  Roll on, two-stage module
1314            delete... */
1315         synchronize_net();
1316
1317         nf_ct_event_cache_flush();
1318  i_see_dead_people:
1319         nf_conntrack_flush();
1320         if (atomic_read(&nf_conntrack_count) != 0) {
1321                 schedule();
1322                 goto i_see_dead_people;
1323         }
1324         /* wait until all references to nf_conntrack_untracked are dropped */
1325         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1326                 schedule();
1327
1328         for (i = 0; i < NF_CT_F_NUM; i++) {
1329                 if (nf_ct_cache[i].use == 0)
1330                         continue;
1331
1332                 NF_CT_ASSERT(nf_ct_cache[i].use == 1);
1333                 nf_ct_cache[i].use = 1;
1334                 nf_conntrack_unregister_cache(i);
1335         }
1336         kmem_cache_destroy(nf_conntrack_expect_cachep);
1337         free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
1338                             nf_conntrack_htable_size);
1339
1340         /* free l3proto protocol tables */
1341         for (i = 0; i < PF_MAX; i++)
1342                 if (nf_ct_protos[i]) {
1343                         kfree(nf_ct_protos[i]);
1344                         nf_ct_protos[i] = NULL;
1345                 }
1346 }
1347
1348 static struct list_head *alloc_hashtable(int size, int *vmalloced)
1349 {
1350         struct list_head *hash;
1351         unsigned int i;
1352
1353         *vmalloced = 0;
1354         hash = (void*)__get_free_pages(GFP_KERNEL,
1355                                        get_order(sizeof(struct list_head)
1356                                                  * size));
1357         if (!hash) {
1358                 *vmalloced = 1;
1359                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1360                 hash = vmalloc(sizeof(struct list_head) * size);
1361         }
1362
1363         if (hash)
1364                 for (i = 0; i < size; i++)
1365                         INIT_LIST_HEAD(&hash[i]);
1366
1367         return hash;
1368 }
1369
1370 int set_hashsize(const char *val, struct kernel_param *kp)
1371 {
1372         int i, bucket, hashsize, vmalloced;
1373         int old_vmalloced, old_size;
1374         int rnd;
1375         struct list_head *hash, *old_hash;
1376         struct nf_conntrack_tuple_hash *h;
1377
1378         /* On boot, we can set this without any fancy locking. */
1379         if (!nf_conntrack_htable_size)
1380                 return param_set_uint(val, kp);
1381
1382         hashsize = simple_strtol(val, NULL, 0);
1383         if (!hashsize)
1384                 return -EINVAL;
1385
1386         hash = alloc_hashtable(hashsize, &vmalloced);
1387         if (!hash)
1388                 return -ENOMEM;
1389
1390         /* We have to rehahs for the new table anyway, so we also can
1391          * use a newrandom seed */
1392         get_random_bytes(&rnd, 4);
1393
1394         write_lock_bh(&nf_conntrack_lock);
1395         for (i = 0; i < nf_conntrack_htable_size; i++) {
1396                 while (!list_empty(&nf_conntrack_hash[i])) {
1397                         h = list_entry(nf_conntrack_hash[i].next,
1398                                        struct nf_conntrack_tuple_hash, list);
1399                         list_del(&h->list);
1400                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1401                         list_add_tail(&h->list, &hash[bucket]);
1402                 }
1403         }
1404         old_size = nf_conntrack_htable_size;
1405         old_vmalloced = nf_conntrack_vmalloc;
1406         old_hash = nf_conntrack_hash;
1407
1408         nf_conntrack_htable_size = hashsize;
1409         nf_conntrack_vmalloc = vmalloced;
1410         nf_conntrack_hash = hash;
1411         nf_conntrack_hash_rnd = rnd;
1412         write_unlock_bh(&nf_conntrack_lock);
1413
1414         free_conntrack_hash(old_hash, old_vmalloced, old_size);
1415         return 0;
1416 }
1417
1418 module_param_call(hashsize, set_hashsize, param_get_uint,
1419                   &nf_conntrack_htable_size, 0600);
1420
1421 int __init nf_conntrack_init(void)
1422 {
1423         unsigned int i;
1424         int ret;
1425
1426         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1427          * machine has 256 buckets.  >= 1GB machines have 8192 buckets. */
1428         if (!nf_conntrack_htable_size) {
1429                 nf_conntrack_htable_size
1430                         = (((num_physpages << PAGE_SHIFT) / 16384)
1431                            / sizeof(struct list_head));
1432                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1433                         nf_conntrack_htable_size = 8192;
1434                 if (nf_conntrack_htable_size < 16)
1435                         nf_conntrack_htable_size = 16;
1436         }
1437         nf_conntrack_max = 8 * nf_conntrack_htable_size;
1438
1439         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1440                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1441                nf_conntrack_max);
1442
1443         nf_conntrack_hash = alloc_hashtable(nf_conntrack_htable_size,
1444                                             &nf_conntrack_vmalloc);
1445         if (!nf_conntrack_hash) {
1446                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1447                 goto err_out;
1448         }
1449
1450         ret = nf_conntrack_register_cache(NF_CT_F_BASIC, "nf_conntrack:basic",
1451                                           sizeof(struct nf_conn));
1452         if (ret < 0) {
1453                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1454                 goto err_free_hash;
1455         }
1456
1457         nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect",
1458                                         sizeof(struct nf_conntrack_expect),
1459                                         0, 0, NULL, NULL);
1460         if (!nf_conntrack_expect_cachep) {
1461                 printk(KERN_ERR "Unable to create nf_expect slab cache\n");
1462                 goto err_free_conntrack_slab;
1463         }
1464
1465         /* Don't NEED lock here, but good form anyway. */
1466         write_lock_bh(&nf_conntrack_lock);
1467         for (i = 0; i < PF_MAX; i++)
1468                 nf_ct_l3protos[i] = &nf_conntrack_generic_l3proto;
1469         write_unlock_bh(&nf_conntrack_lock);
1470
1471         /* For use by REJECT target */
1472         ip_ct_attach = __nf_conntrack_attach;
1473
1474         /* Set up fake conntrack:
1475             - to never be deleted, not in any hashes */
1476         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1477         /*  - and look it like as a confirmed connection */
1478         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1479
1480         return ret;
1481
1482 err_free_conntrack_slab:
1483         nf_conntrack_unregister_cache(NF_CT_F_BASIC);
1484 err_free_hash:
1485         free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
1486                             nf_conntrack_htable_size);
1487 err_out:
1488         return -ENOMEM;
1489 }