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 * (C) 2005-2012 Patrick McHardy <kaber@trash.net> 9 * 10 * This program is free software; you can redistribute it and/or modify 11 * it under the terms of the GNU General Public License version 2 as 12 * published by the Free Software Foundation. 13 */ 14 15 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 16 17 #include <linux/types.h> 18 #include <linux/netfilter.h> 19 #include <linux/module.h> 20 #include <linux/sched.h> 21 #include <linux/skbuff.h> 22 #include <linux/proc_fs.h> 23 #include <linux/vmalloc.h> 24 #include <linux/stddef.h> 25 #include <linux/slab.h> 26 #include <linux/random.h> 27 #include <linux/jhash.h> 28 #include <linux/err.h> 29 #include <linux/percpu.h> 30 #include <linux/moduleparam.h> 31 #include <linux/notifier.h> 32 #include <linux/kernel.h> 33 #include <linux/netdevice.h> 34 #include <linux/socket.h> 35 #include <linux/mm.h> 36 #include <linux/nsproxy.h> 37 #include <linux/rculist_nulls.h> 38 39 #include <net/netfilter/nf_conntrack.h> 40 #include <net/netfilter/nf_conntrack_l3proto.h> 41 #include <net/netfilter/nf_conntrack_l4proto.h> 42 #include <net/netfilter/nf_conntrack_expect.h> 43 #include <net/netfilter/nf_conntrack_helper.h> 44 #include <net/netfilter/nf_conntrack_seqadj.h> 45 #include <net/netfilter/nf_conntrack_core.h> 46 #include <net/netfilter/nf_conntrack_extend.h> 47 #include <net/netfilter/nf_conntrack_acct.h> 48 #include <net/netfilter/nf_conntrack_ecache.h> 49 #include <net/netfilter/nf_conntrack_zones.h> 50 #include <net/netfilter/nf_conntrack_timestamp.h> 51 #include <net/netfilter/nf_conntrack_timeout.h> 52 #include <net/netfilter/nf_conntrack_labels.h> 53 #include <net/netfilter/nf_conntrack_synproxy.h> 54 #include <net/netfilter/nf_nat.h> 55 #include <net/netfilter/nf_nat_core.h> 56 #include <net/netfilter/nf_nat_helper.h> 57 #include <net/netns/hash.h> 58 59 #define NF_CONNTRACK_VERSION "0.5.0" 60 61 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct, 62 enum nf_nat_manip_type manip, 63 const struct nlattr *attr) __read_mostly; 64 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook); 65 66 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS]; 67 EXPORT_SYMBOL_GPL(nf_conntrack_locks); 68 69 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock); 70 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock); 71 72 struct hlist_nulls_head *nf_conntrack_hash __read_mostly; 73 EXPORT_SYMBOL_GPL(nf_conntrack_hash); 74 75 static __read_mostly struct kmem_cache *nf_conntrack_cachep; 76 static __read_mostly spinlock_t nf_conntrack_locks_all_lock; 77 static __read_mostly seqcount_t nf_conntrack_generation; 78 static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock); 79 static __read_mostly bool nf_conntrack_locks_all; 80 81 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock) 82 { 83 spin_lock(lock); 84 while (unlikely(nf_conntrack_locks_all)) { 85 spin_unlock(lock); 86 spin_unlock_wait(&nf_conntrack_locks_all_lock); 87 spin_lock(lock); 88 } 89 } 90 EXPORT_SYMBOL_GPL(nf_conntrack_lock); 91 92 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2) 93 { 94 h1 %= CONNTRACK_LOCKS; 95 h2 %= CONNTRACK_LOCKS; 96 spin_unlock(&nf_conntrack_locks[h1]); 97 if (h1 != h2) 98 spin_unlock(&nf_conntrack_locks[h2]); 99 } 100 101 /* return true if we need to recompute hashes (in case hash table was resized) */ 102 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1, 103 unsigned int h2, unsigned int sequence) 104 { 105 h1 %= CONNTRACK_LOCKS; 106 h2 %= CONNTRACK_LOCKS; 107 if (h1 <= h2) { 108 nf_conntrack_lock(&nf_conntrack_locks[h1]); 109 if (h1 != h2) 110 spin_lock_nested(&nf_conntrack_locks[h2], 111 SINGLE_DEPTH_NESTING); 112 } else { 113 nf_conntrack_lock(&nf_conntrack_locks[h2]); 114 spin_lock_nested(&nf_conntrack_locks[h1], 115 SINGLE_DEPTH_NESTING); 116 } 117 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) { 118 nf_conntrack_double_unlock(h1, h2); 119 return true; 120 } 121 return false; 122 } 123 124 static void nf_conntrack_all_lock(void) 125 { 126 int i; 127 128 spin_lock(&nf_conntrack_locks_all_lock); 129 nf_conntrack_locks_all = true; 130 131 for (i = 0; i < CONNTRACK_LOCKS; i++) { 132 spin_unlock_wait(&nf_conntrack_locks[i]); 133 } 134 } 135 136 static void nf_conntrack_all_unlock(void) 137 { 138 nf_conntrack_locks_all = false; 139 spin_unlock(&nf_conntrack_locks_all_lock); 140 } 141 142 unsigned int nf_conntrack_htable_size __read_mostly; 143 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); 144 145 unsigned int nf_conntrack_max __read_mostly; 146 EXPORT_SYMBOL_GPL(nf_conntrack_max); 147 148 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked); 149 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked); 150 151 static unsigned int nf_conntrack_hash_rnd __read_mostly; 152 153 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, 154 const struct net *net) 155 { 156 unsigned int n; 157 u32 seed; 158 159 get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd)); 160 161 /* The direction must be ignored, so we hash everything up to the 162 * destination ports (which is a multiple of 4) and treat the last 163 * three bytes manually. 164 */ 165 seed = nf_conntrack_hash_rnd ^ net_hash_mix(net); 166 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); 167 return jhash2((u32 *)tuple, n, seed ^ 168 (((__force __u16)tuple->dst.u.all << 16) | 169 tuple->dst.protonum)); 170 } 171 172 static u32 scale_hash(u32 hash) 173 { 174 return reciprocal_scale(hash, nf_conntrack_htable_size); 175 } 176 177 static u32 __hash_conntrack(const struct net *net, 178 const struct nf_conntrack_tuple *tuple, 179 unsigned int size) 180 { 181 return reciprocal_scale(hash_conntrack_raw(tuple, net), size); 182 } 183 184 static u32 hash_conntrack(const struct net *net, 185 const struct nf_conntrack_tuple *tuple) 186 { 187 return scale_hash(hash_conntrack_raw(tuple, net)); 188 } 189 190 bool 191 nf_ct_get_tuple(const struct sk_buff *skb, 192 unsigned int nhoff, 193 unsigned int dataoff, 194 u_int16_t l3num, 195 u_int8_t protonum, 196 struct net *net, 197 struct nf_conntrack_tuple *tuple, 198 const struct nf_conntrack_l3proto *l3proto, 199 const struct nf_conntrack_l4proto *l4proto) 200 { 201 memset(tuple, 0, sizeof(*tuple)); 202 203 tuple->src.l3num = l3num; 204 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) 205 return false; 206 207 tuple->dst.protonum = protonum; 208 tuple->dst.dir = IP_CT_DIR_ORIGINAL; 209 210 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple); 211 } 212 EXPORT_SYMBOL_GPL(nf_ct_get_tuple); 213 214 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, 215 u_int16_t l3num, 216 struct net *net, struct nf_conntrack_tuple *tuple) 217 { 218 struct nf_conntrack_l3proto *l3proto; 219 struct nf_conntrack_l4proto *l4proto; 220 unsigned int protoff; 221 u_int8_t protonum; 222 int ret; 223 224 rcu_read_lock(); 225 226 l3proto = __nf_ct_l3proto_find(l3num); 227 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); 228 if (ret != NF_ACCEPT) { 229 rcu_read_unlock(); 230 return false; 231 } 232 233 l4proto = __nf_ct_l4proto_find(l3num, protonum); 234 235 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple, 236 l3proto, l4proto); 237 238 rcu_read_unlock(); 239 return ret; 240 } 241 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); 242 243 bool 244 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, 245 const struct nf_conntrack_tuple *orig, 246 const struct nf_conntrack_l3proto *l3proto, 247 const struct nf_conntrack_l4proto *l4proto) 248 { 249 memset(inverse, 0, sizeof(*inverse)); 250 251 inverse->src.l3num = orig->src.l3num; 252 if (l3proto->invert_tuple(inverse, orig) == 0) 253 return false; 254 255 inverse->dst.dir = !orig->dst.dir; 256 257 inverse->dst.protonum = orig->dst.protonum; 258 return l4proto->invert_tuple(inverse, orig); 259 } 260 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); 261 262 static void 263 clean_from_lists(struct nf_conn *ct) 264 { 265 pr_debug("clean_from_lists(%p)\n", ct); 266 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); 267 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); 268 269 /* Destroy all pending expectations */ 270 nf_ct_remove_expectations(ct); 271 } 272 273 /* must be called with local_bh_disable */ 274 static void nf_ct_add_to_dying_list(struct nf_conn *ct) 275 { 276 struct ct_pcpu *pcpu; 277 278 /* add this conntrack to the (per cpu) dying list */ 279 ct->cpu = smp_processor_id(); 280 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); 281 282 spin_lock(&pcpu->lock); 283 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, 284 &pcpu->dying); 285 spin_unlock(&pcpu->lock); 286 } 287 288 /* must be called with local_bh_disable */ 289 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct) 290 { 291 struct ct_pcpu *pcpu; 292 293 /* add this conntrack to the (per cpu) unconfirmed list */ 294 ct->cpu = smp_processor_id(); 295 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); 296 297 spin_lock(&pcpu->lock); 298 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, 299 &pcpu->unconfirmed); 300 spin_unlock(&pcpu->lock); 301 } 302 303 /* must be called with local_bh_disable */ 304 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct) 305 { 306 struct ct_pcpu *pcpu; 307 308 /* We overload first tuple to link into unconfirmed or dying list.*/ 309 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu); 310 311 spin_lock(&pcpu->lock); 312 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); 313 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); 314 spin_unlock(&pcpu->lock); 315 } 316 317 /* Released via destroy_conntrack() */ 318 struct nf_conn *nf_ct_tmpl_alloc(struct net *net, 319 const struct nf_conntrack_zone *zone, 320 gfp_t flags) 321 { 322 struct nf_conn *tmpl; 323 324 tmpl = kzalloc(sizeof(*tmpl), flags); 325 if (tmpl == NULL) 326 return NULL; 327 328 tmpl->status = IPS_TEMPLATE; 329 write_pnet(&tmpl->ct_net, net); 330 331 if (nf_ct_zone_add(tmpl, flags, zone) < 0) 332 goto out_free; 333 334 atomic_set(&tmpl->ct_general.use, 0); 335 336 return tmpl; 337 out_free: 338 kfree(tmpl); 339 return NULL; 340 } 341 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc); 342 343 void nf_ct_tmpl_free(struct nf_conn *tmpl) 344 { 345 nf_ct_ext_destroy(tmpl); 346 nf_ct_ext_free(tmpl); 347 kfree(tmpl); 348 } 349 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free); 350 351 static void 352 destroy_conntrack(struct nf_conntrack *nfct) 353 { 354 struct nf_conn *ct = (struct nf_conn *)nfct; 355 struct net *net = nf_ct_net(ct); 356 struct nf_conntrack_l4proto *l4proto; 357 358 pr_debug("destroy_conntrack(%p)\n", ct); 359 NF_CT_ASSERT(atomic_read(&nfct->use) == 0); 360 NF_CT_ASSERT(!timer_pending(&ct->timeout)); 361 362 if (unlikely(nf_ct_is_template(ct))) { 363 nf_ct_tmpl_free(ct); 364 return; 365 } 366 rcu_read_lock(); 367 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); 368 if (l4proto->destroy) 369 l4proto->destroy(ct); 370 371 rcu_read_unlock(); 372 373 local_bh_disable(); 374 /* Expectations will have been removed in clean_from_lists, 375 * except TFTP can create an expectation on the first packet, 376 * before connection is in the list, so we need to clean here, 377 * too. 378 */ 379 nf_ct_remove_expectations(ct); 380 381 nf_ct_del_from_dying_or_unconfirmed_list(ct); 382 383 NF_CT_STAT_INC(net, delete); 384 local_bh_enable(); 385 386 if (ct->master) 387 nf_ct_put(ct->master); 388 389 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); 390 nf_conntrack_free(ct); 391 } 392 393 static void nf_ct_delete_from_lists(struct nf_conn *ct) 394 { 395 struct net *net = nf_ct_net(ct); 396 unsigned int hash, reply_hash; 397 unsigned int sequence; 398 399 nf_ct_helper_destroy(ct); 400 401 local_bh_disable(); 402 do { 403 sequence = read_seqcount_begin(&nf_conntrack_generation); 404 hash = hash_conntrack(net, 405 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 406 reply_hash = hash_conntrack(net, 407 &ct->tuplehash[IP_CT_DIR_REPLY].tuple); 408 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); 409 410 clean_from_lists(ct); 411 nf_conntrack_double_unlock(hash, reply_hash); 412 413 nf_ct_add_to_dying_list(ct); 414 415 NF_CT_STAT_INC(net, delete_list); 416 local_bh_enable(); 417 } 418 419 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report) 420 { 421 struct nf_conn_tstamp *tstamp; 422 423 tstamp = nf_conn_tstamp_find(ct); 424 if (tstamp && tstamp->stop == 0) 425 tstamp->stop = ktime_get_real_ns(); 426 427 if (nf_ct_is_dying(ct)) 428 goto delete; 429 430 if (nf_conntrack_event_report(IPCT_DESTROY, ct, 431 portid, report) < 0) { 432 /* destroy event was not delivered */ 433 nf_ct_delete_from_lists(ct); 434 nf_conntrack_ecache_delayed_work(nf_ct_net(ct)); 435 return false; 436 } 437 438 nf_conntrack_ecache_work(nf_ct_net(ct)); 439 set_bit(IPS_DYING_BIT, &ct->status); 440 delete: 441 nf_ct_delete_from_lists(ct); 442 nf_ct_put(ct); 443 return true; 444 } 445 EXPORT_SYMBOL_GPL(nf_ct_delete); 446 447 static void death_by_timeout(unsigned long ul_conntrack) 448 { 449 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0); 450 } 451 452 static inline bool 453 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h, 454 const struct nf_conntrack_tuple *tuple, 455 const struct nf_conntrack_zone *zone, 456 const struct net *net) 457 { 458 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); 459 460 /* A conntrack can be recreated with the equal tuple, 461 * so we need to check that the conntrack is confirmed 462 */ 463 return nf_ct_tuple_equal(tuple, &h->tuple) && 464 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) && 465 nf_ct_is_confirmed(ct) && 466 net_eq(net, nf_ct_net(ct)); 467 } 468 469 /* 470 * Warning : 471 * - Caller must take a reference on returned object 472 * and recheck nf_ct_tuple_equal(tuple, &h->tuple) 473 */ 474 static struct nf_conntrack_tuple_hash * 475 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone, 476 const struct nf_conntrack_tuple *tuple, u32 hash) 477 { 478 struct nf_conntrack_tuple_hash *h; 479 struct hlist_nulls_head *ct_hash; 480 struct hlist_nulls_node *n; 481 unsigned int bucket, sequence; 482 483 begin: 484 do { 485 sequence = read_seqcount_begin(&nf_conntrack_generation); 486 bucket = scale_hash(hash); 487 ct_hash = nf_conntrack_hash; 488 } while (read_seqcount_retry(&nf_conntrack_generation, sequence)); 489 490 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) { 491 if (nf_ct_key_equal(h, tuple, zone, net)) { 492 NF_CT_STAT_INC_ATOMIC(net, found); 493 return h; 494 } 495 NF_CT_STAT_INC_ATOMIC(net, searched); 496 } 497 /* 498 * if the nulls value we got at the end of this lookup is 499 * not the expected one, we must restart lookup. 500 * We probably met an item that was moved to another chain. 501 */ 502 if (get_nulls_value(n) != bucket) { 503 NF_CT_STAT_INC_ATOMIC(net, search_restart); 504 goto begin; 505 } 506 507 return NULL; 508 } 509 510 /* Find a connection corresponding to a tuple. */ 511 static struct nf_conntrack_tuple_hash * 512 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, 513 const struct nf_conntrack_tuple *tuple, u32 hash) 514 { 515 struct nf_conntrack_tuple_hash *h; 516 struct nf_conn *ct; 517 518 rcu_read_lock(); 519 begin: 520 h = ____nf_conntrack_find(net, zone, tuple, hash); 521 if (h) { 522 ct = nf_ct_tuplehash_to_ctrack(h); 523 if (unlikely(nf_ct_is_dying(ct) || 524 !atomic_inc_not_zero(&ct->ct_general.use))) 525 h = NULL; 526 else { 527 if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) { 528 nf_ct_put(ct); 529 goto begin; 530 } 531 } 532 } 533 rcu_read_unlock(); 534 535 return h; 536 } 537 538 struct nf_conntrack_tuple_hash * 539 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone, 540 const struct nf_conntrack_tuple *tuple) 541 { 542 return __nf_conntrack_find_get(net, zone, tuple, 543 hash_conntrack_raw(tuple, net)); 544 } 545 EXPORT_SYMBOL_GPL(nf_conntrack_find_get); 546 547 static void __nf_conntrack_hash_insert(struct nf_conn *ct, 548 unsigned int hash, 549 unsigned int reply_hash) 550 { 551 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, 552 &nf_conntrack_hash[hash]); 553 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, 554 &nf_conntrack_hash[reply_hash]); 555 } 556 557 int 558 nf_conntrack_hash_check_insert(struct nf_conn *ct) 559 { 560 const struct nf_conntrack_zone *zone; 561 struct net *net = nf_ct_net(ct); 562 unsigned int hash, reply_hash; 563 struct nf_conntrack_tuple_hash *h; 564 struct hlist_nulls_node *n; 565 unsigned int sequence; 566 567 zone = nf_ct_zone(ct); 568 569 local_bh_disable(); 570 do { 571 sequence = read_seqcount_begin(&nf_conntrack_generation); 572 hash = hash_conntrack(net, 573 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 574 reply_hash = hash_conntrack(net, 575 &ct->tuplehash[IP_CT_DIR_REPLY].tuple); 576 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); 577 578 /* See if there's one in the list already, including reverse */ 579 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) 580 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, 581 zone, net)) 582 goto out; 583 584 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) 585 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, 586 zone, net)) 587 goto out; 588 589 add_timer(&ct->timeout); 590 smp_wmb(); 591 /* The caller holds a reference to this object */ 592 atomic_set(&ct->ct_general.use, 2); 593 __nf_conntrack_hash_insert(ct, hash, reply_hash); 594 nf_conntrack_double_unlock(hash, reply_hash); 595 NF_CT_STAT_INC(net, insert); 596 local_bh_enable(); 597 return 0; 598 599 out: 600 nf_conntrack_double_unlock(hash, reply_hash); 601 NF_CT_STAT_INC(net, insert_failed); 602 local_bh_enable(); 603 return -EEXIST; 604 } 605 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert); 606 607 static inline void nf_ct_acct_update(struct nf_conn *ct, 608 enum ip_conntrack_info ctinfo, 609 unsigned int len) 610 { 611 struct nf_conn_acct *acct; 612 613 acct = nf_conn_acct_find(ct); 614 if (acct) { 615 struct nf_conn_counter *counter = acct->counter; 616 617 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets); 618 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes); 619 } 620 } 621 622 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo, 623 const struct nf_conn *loser_ct) 624 { 625 struct nf_conn_acct *acct; 626 627 acct = nf_conn_acct_find(loser_ct); 628 if (acct) { 629 struct nf_conn_counter *counter = acct->counter; 630 unsigned int bytes; 631 632 /* u32 should be fine since we must have seen one packet. */ 633 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes); 634 nf_ct_acct_update(ct, ctinfo, bytes); 635 } 636 } 637 638 /* Resolve race on insertion if this protocol allows this. */ 639 static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb, 640 enum ip_conntrack_info ctinfo, 641 struct nf_conntrack_tuple_hash *h) 642 { 643 /* This is the conntrack entry already in hashes that won race. */ 644 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); 645 struct nf_conntrack_l4proto *l4proto; 646 647 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); 648 if (l4proto->allow_clash && 649 !nfct_nat(ct) && 650 !nf_ct_is_dying(ct) && 651 atomic_inc_not_zero(&ct->ct_general.use)) { 652 nf_ct_acct_merge(ct, ctinfo, (struct nf_conn *)skb->nfct); 653 nf_conntrack_put(skb->nfct); 654 /* Assign conntrack already in hashes to this skbuff. Don't 655 * modify skb->nfctinfo to ensure consistent stateful filtering. 656 */ 657 skb->nfct = &ct->ct_general; 658 return NF_ACCEPT; 659 } 660 NF_CT_STAT_INC(net, drop); 661 return NF_DROP; 662 } 663 664 /* Confirm a connection given skb; places it in hash table */ 665 int 666 __nf_conntrack_confirm(struct sk_buff *skb) 667 { 668 const struct nf_conntrack_zone *zone; 669 unsigned int hash, reply_hash; 670 struct nf_conntrack_tuple_hash *h; 671 struct nf_conn *ct; 672 struct nf_conn_help *help; 673 struct nf_conn_tstamp *tstamp; 674 struct hlist_nulls_node *n; 675 enum ip_conntrack_info ctinfo; 676 struct net *net; 677 unsigned int sequence; 678 int ret = NF_DROP; 679 680 ct = nf_ct_get(skb, &ctinfo); 681 net = nf_ct_net(ct); 682 683 /* ipt_REJECT uses nf_conntrack_attach to attach related 684 ICMP/TCP RST packets in other direction. Actual packet 685 which created connection will be IP_CT_NEW or for an 686 expected connection, IP_CT_RELATED. */ 687 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) 688 return NF_ACCEPT; 689 690 zone = nf_ct_zone(ct); 691 local_bh_disable(); 692 693 do { 694 sequence = read_seqcount_begin(&nf_conntrack_generation); 695 /* reuse the hash saved before */ 696 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; 697 hash = scale_hash(hash); 698 reply_hash = hash_conntrack(net, 699 &ct->tuplehash[IP_CT_DIR_REPLY].tuple); 700 701 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence)); 702 703 /* We're not in hash table, and we refuse to set up related 704 * connections for unconfirmed conns. But packet copies and 705 * REJECT will give spurious warnings here. 706 */ 707 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ 708 709 /* No external references means no one else could have 710 * confirmed us. 711 */ 712 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 713 pr_debug("Confirming conntrack %p\n", ct); 714 /* We have to check the DYING flag after unlink to prevent 715 * a race against nf_ct_get_next_corpse() possibly called from 716 * user context, else we insert an already 'dead' hash, blocking 717 * further use of that particular connection -JM. 718 */ 719 nf_ct_del_from_dying_or_unconfirmed_list(ct); 720 721 if (unlikely(nf_ct_is_dying(ct))) { 722 nf_ct_add_to_dying_list(ct); 723 goto dying; 724 } 725 726 /* See if there's one in the list already, including reverse: 727 NAT could have grabbed it without realizing, since we're 728 not in the hash. If there is, we lost race. */ 729 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) 730 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, 731 zone, net)) 732 goto out; 733 734 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) 735 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple, 736 zone, net)) 737 goto out; 738 739 /* Timer relative to confirmation time, not original 740 setting time, otherwise we'd get timer wrap in 741 weird delay cases. */ 742 ct->timeout.expires += jiffies; 743 add_timer(&ct->timeout); 744 atomic_inc(&ct->ct_general.use); 745 ct->status |= IPS_CONFIRMED; 746 747 /* set conntrack timestamp, if enabled. */ 748 tstamp = nf_conn_tstamp_find(ct); 749 if (tstamp) { 750 if (skb->tstamp.tv64 == 0) 751 __net_timestamp(skb); 752 753 tstamp->start = ktime_to_ns(skb->tstamp); 754 } 755 /* Since the lookup is lockless, hash insertion must be done after 756 * starting the timer and setting the CONFIRMED bit. The RCU barriers 757 * guarantee that no other CPU can find the conntrack before the above 758 * stores are visible. 759 */ 760 __nf_conntrack_hash_insert(ct, hash, reply_hash); 761 nf_conntrack_double_unlock(hash, reply_hash); 762 NF_CT_STAT_INC(net, insert); 763 local_bh_enable(); 764 765 help = nfct_help(ct); 766 if (help && help->helper) 767 nf_conntrack_event_cache(IPCT_HELPER, ct); 768 769 nf_conntrack_event_cache(master_ct(ct) ? 770 IPCT_RELATED : IPCT_NEW, ct); 771 return NF_ACCEPT; 772 773 out: 774 nf_ct_add_to_dying_list(ct); 775 ret = nf_ct_resolve_clash(net, skb, ctinfo, h); 776 dying: 777 nf_conntrack_double_unlock(hash, reply_hash); 778 NF_CT_STAT_INC(net, insert_failed); 779 local_bh_enable(); 780 return ret; 781 } 782 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); 783 784 /* Returns true if a connection correspondings to the tuple (required 785 for NAT). */ 786 int 787 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, 788 const struct nf_conn *ignored_conntrack) 789 { 790 struct net *net = nf_ct_net(ignored_conntrack); 791 const struct nf_conntrack_zone *zone; 792 struct nf_conntrack_tuple_hash *h; 793 struct hlist_nulls_head *ct_hash; 794 unsigned int hash, sequence; 795 struct hlist_nulls_node *n; 796 struct nf_conn *ct; 797 798 zone = nf_ct_zone(ignored_conntrack); 799 800 rcu_read_lock(); 801 do { 802 sequence = read_seqcount_begin(&nf_conntrack_generation); 803 hash = hash_conntrack(net, tuple); 804 ct_hash = nf_conntrack_hash; 805 } while (read_seqcount_retry(&nf_conntrack_generation, sequence)); 806 807 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) { 808 ct = nf_ct_tuplehash_to_ctrack(h); 809 if (ct != ignored_conntrack && 810 nf_ct_key_equal(h, tuple, zone, net)) { 811 NF_CT_STAT_INC_ATOMIC(net, found); 812 rcu_read_unlock(); 813 return 1; 814 } 815 NF_CT_STAT_INC_ATOMIC(net, searched); 816 } 817 rcu_read_unlock(); 818 819 return 0; 820 } 821 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); 822 823 #define NF_CT_EVICTION_RANGE 8 824 825 /* There's a small race here where we may free a just-assured 826 connection. Too bad: we're in trouble anyway. */ 827 static noinline int early_drop(struct net *net, unsigned int _hash) 828 { 829 /* Use oldest entry, which is roughly LRU */ 830 struct nf_conntrack_tuple_hash *h; 831 struct nf_conn *tmp; 832 struct hlist_nulls_node *n; 833 unsigned int i, hash, sequence; 834 struct nf_conn *ct = NULL; 835 spinlock_t *lockp; 836 bool ret = false; 837 838 i = 0; 839 840 local_bh_disable(); 841 restart: 842 sequence = read_seqcount_begin(&nf_conntrack_generation); 843 for (; i < NF_CT_EVICTION_RANGE; i++) { 844 hash = scale_hash(_hash++); 845 lockp = &nf_conntrack_locks[hash % CONNTRACK_LOCKS]; 846 nf_conntrack_lock(lockp); 847 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) { 848 spin_unlock(lockp); 849 goto restart; 850 } 851 hlist_nulls_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], 852 hnnode) { 853 tmp = nf_ct_tuplehash_to_ctrack(h); 854 855 if (test_bit(IPS_ASSURED_BIT, &tmp->status) || 856 !net_eq(nf_ct_net(tmp), net) || 857 nf_ct_is_dying(tmp)) 858 continue; 859 860 if (atomic_inc_not_zero(&tmp->ct_general.use)) { 861 ct = tmp; 862 break; 863 } 864 } 865 866 spin_unlock(lockp); 867 if (ct) 868 break; 869 } 870 871 local_bh_enable(); 872 873 if (!ct) 874 return false; 875 876 /* kill only if in same netns -- might have moved due to 877 * SLAB_DESTROY_BY_RCU rules 878 */ 879 if (net_eq(nf_ct_net(ct), net) && del_timer(&ct->timeout)) { 880 if (nf_ct_delete(ct, 0, 0)) { 881 NF_CT_STAT_INC_ATOMIC(net, early_drop); 882 ret = true; 883 } 884 } 885 886 nf_ct_put(ct); 887 return ret; 888 } 889 890 static struct nf_conn * 891 __nf_conntrack_alloc(struct net *net, 892 const struct nf_conntrack_zone *zone, 893 const struct nf_conntrack_tuple *orig, 894 const struct nf_conntrack_tuple *repl, 895 gfp_t gfp, u32 hash) 896 { 897 struct nf_conn *ct; 898 899 /* We don't want any race condition at early drop stage */ 900 atomic_inc(&net->ct.count); 901 902 if (nf_conntrack_max && 903 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { 904 if (!early_drop(net, hash)) { 905 atomic_dec(&net->ct.count); 906 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n"); 907 return ERR_PTR(-ENOMEM); 908 } 909 } 910 911 /* 912 * Do not use kmem_cache_zalloc(), as this cache uses 913 * SLAB_DESTROY_BY_RCU. 914 */ 915 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp); 916 if (ct == NULL) 917 goto out; 918 919 spin_lock_init(&ct->lock); 920 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; 921 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; 922 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; 923 /* save hash for reusing when confirming */ 924 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; 925 ct->status = 0; 926 /* Don't set timer yet: wait for confirmation */ 927 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); 928 write_pnet(&ct->ct_net, net); 929 memset(&ct->__nfct_init_offset[0], 0, 930 offsetof(struct nf_conn, proto) - 931 offsetof(struct nf_conn, __nfct_init_offset[0])); 932 933 if (zone && nf_ct_zone_add(ct, GFP_ATOMIC, zone) < 0) 934 goto out_free; 935 936 /* Because we use RCU lookups, we set ct_general.use to zero before 937 * this is inserted in any list. 938 */ 939 atomic_set(&ct->ct_general.use, 0); 940 return ct; 941 out_free: 942 kmem_cache_free(nf_conntrack_cachep, ct); 943 out: 944 atomic_dec(&net->ct.count); 945 return ERR_PTR(-ENOMEM); 946 } 947 948 struct nf_conn *nf_conntrack_alloc(struct net *net, 949 const struct nf_conntrack_zone *zone, 950 const struct nf_conntrack_tuple *orig, 951 const struct nf_conntrack_tuple *repl, 952 gfp_t gfp) 953 { 954 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); 955 } 956 EXPORT_SYMBOL_GPL(nf_conntrack_alloc); 957 958 void nf_conntrack_free(struct nf_conn *ct) 959 { 960 struct net *net = nf_ct_net(ct); 961 962 /* A freed object has refcnt == 0, that's 963 * the golden rule for SLAB_DESTROY_BY_RCU 964 */ 965 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0); 966 967 nf_ct_ext_destroy(ct); 968 nf_ct_ext_free(ct); 969 kmem_cache_free(nf_conntrack_cachep, ct); 970 smp_mb__before_atomic(); 971 atomic_dec(&net->ct.count); 972 } 973 EXPORT_SYMBOL_GPL(nf_conntrack_free); 974 975 976 /* Allocate a new conntrack: we return -ENOMEM if classification 977 failed due to stress. Otherwise it really is unclassifiable. */ 978 static struct nf_conntrack_tuple_hash * 979 init_conntrack(struct net *net, struct nf_conn *tmpl, 980 const struct nf_conntrack_tuple *tuple, 981 struct nf_conntrack_l3proto *l3proto, 982 struct nf_conntrack_l4proto *l4proto, 983 struct sk_buff *skb, 984 unsigned int dataoff, u32 hash) 985 { 986 struct nf_conn *ct; 987 struct nf_conn_help *help; 988 struct nf_conntrack_tuple repl_tuple; 989 struct nf_conntrack_ecache *ecache; 990 struct nf_conntrack_expect *exp = NULL; 991 const struct nf_conntrack_zone *zone; 992 struct nf_conn_timeout *timeout_ext; 993 struct nf_conntrack_zone tmp; 994 unsigned int *timeouts; 995 996 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { 997 pr_debug("Can't invert tuple.\n"); 998 return NULL; 999 } 1000 1001 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); 1002 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, 1003 hash); 1004 if (IS_ERR(ct)) 1005 return (struct nf_conntrack_tuple_hash *)ct; 1006 1007 if (tmpl && nfct_synproxy(tmpl)) { 1008 nfct_seqadj_ext_add(ct); 1009 nfct_synproxy_ext_add(ct); 1010 } 1011 1012 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL; 1013 if (timeout_ext) { 1014 timeouts = nf_ct_timeout_data(timeout_ext); 1015 if (unlikely(!timeouts)) 1016 timeouts = l4proto->get_timeouts(net); 1017 } else { 1018 timeouts = l4proto->get_timeouts(net); 1019 } 1020 1021 if (!l4proto->new(ct, skb, dataoff, timeouts)) { 1022 nf_conntrack_free(ct); 1023 pr_debug("can't track with proto module\n"); 1024 return NULL; 1025 } 1026 1027 if (timeout_ext) 1028 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout), 1029 GFP_ATOMIC); 1030 1031 nf_ct_acct_ext_add(ct, GFP_ATOMIC); 1032 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC); 1033 nf_ct_labels_ext_add(ct); 1034 1035 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; 1036 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, 1037 ecache ? ecache->expmask : 0, 1038 GFP_ATOMIC); 1039 1040 local_bh_disable(); 1041 if (net->ct.expect_count) { 1042 spin_lock(&nf_conntrack_expect_lock); 1043 exp = nf_ct_find_expectation(net, zone, tuple); 1044 if (exp) { 1045 pr_debug("expectation arrives ct=%p exp=%p\n", 1046 ct, exp); 1047 /* Welcome, Mr. Bond. We've been expecting you... */ 1048 __set_bit(IPS_EXPECTED_BIT, &ct->status); 1049 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */ 1050 ct->master = exp->master; 1051 if (exp->helper) { 1052 help = nf_ct_helper_ext_add(ct, exp->helper, 1053 GFP_ATOMIC); 1054 if (help) 1055 rcu_assign_pointer(help->helper, exp->helper); 1056 } 1057 1058 #ifdef CONFIG_NF_CONNTRACK_MARK 1059 ct->mark = exp->master->mark; 1060 #endif 1061 #ifdef CONFIG_NF_CONNTRACK_SECMARK 1062 ct->secmark = exp->master->secmark; 1063 #endif 1064 NF_CT_STAT_INC(net, expect_new); 1065 } 1066 spin_unlock(&nf_conntrack_expect_lock); 1067 } 1068 if (!exp) { 1069 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); 1070 NF_CT_STAT_INC(net, new); 1071 } 1072 1073 /* Now it is inserted into the unconfirmed list, bump refcount */ 1074 nf_conntrack_get(&ct->ct_general); 1075 nf_ct_add_to_unconfirmed_list(ct); 1076 1077 local_bh_enable(); 1078 1079 if (exp) { 1080 if (exp->expectfn) 1081 exp->expectfn(ct, exp); 1082 nf_ct_expect_put(exp); 1083 } 1084 1085 return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; 1086 } 1087 1088 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ 1089 static inline struct nf_conn * 1090 resolve_normal_ct(struct net *net, struct nf_conn *tmpl, 1091 struct sk_buff *skb, 1092 unsigned int dataoff, 1093 u_int16_t l3num, 1094 u_int8_t protonum, 1095 struct nf_conntrack_l3proto *l3proto, 1096 struct nf_conntrack_l4proto *l4proto, 1097 int *set_reply, 1098 enum ip_conntrack_info *ctinfo) 1099 { 1100 const struct nf_conntrack_zone *zone; 1101 struct nf_conntrack_tuple tuple; 1102 struct nf_conntrack_tuple_hash *h; 1103 struct nf_conntrack_zone tmp; 1104 struct nf_conn *ct; 1105 u32 hash; 1106 1107 if (!nf_ct_get_tuple(skb, skb_network_offset(skb), 1108 dataoff, l3num, protonum, net, &tuple, l3proto, 1109 l4proto)) { 1110 pr_debug("Can't get tuple\n"); 1111 return NULL; 1112 } 1113 1114 /* look for tuple match */ 1115 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp); 1116 hash = hash_conntrack_raw(&tuple, net); 1117 h = __nf_conntrack_find_get(net, zone, &tuple, hash); 1118 if (!h) { 1119 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, 1120 skb, dataoff, hash); 1121 if (!h) 1122 return NULL; 1123 if (IS_ERR(h)) 1124 return (void *)h; 1125 } 1126 ct = nf_ct_tuplehash_to_ctrack(h); 1127 1128 /* It exists; we have (non-exclusive) reference. */ 1129 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { 1130 *ctinfo = IP_CT_ESTABLISHED_REPLY; 1131 /* Please set reply bit if this packet OK */ 1132 *set_reply = 1; 1133 } else { 1134 /* Once we've had two way comms, always ESTABLISHED. */ 1135 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { 1136 pr_debug("normal packet for %p\n", ct); 1137 *ctinfo = IP_CT_ESTABLISHED; 1138 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { 1139 pr_debug("related packet for %p\n", ct); 1140 *ctinfo = IP_CT_RELATED; 1141 } else { 1142 pr_debug("new packet for %p\n", ct); 1143 *ctinfo = IP_CT_NEW; 1144 } 1145 *set_reply = 0; 1146 } 1147 skb->nfct = &ct->ct_general; 1148 skb->nfctinfo = *ctinfo; 1149 return ct; 1150 } 1151 1152 unsigned int 1153 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, 1154 struct sk_buff *skb) 1155 { 1156 struct nf_conn *ct, *tmpl = NULL; 1157 enum ip_conntrack_info ctinfo; 1158 struct nf_conntrack_l3proto *l3proto; 1159 struct nf_conntrack_l4proto *l4proto; 1160 unsigned int *timeouts; 1161 unsigned int dataoff; 1162 u_int8_t protonum; 1163 int set_reply = 0; 1164 int ret; 1165 1166 if (skb->nfct) { 1167 /* Previously seen (loopback or untracked)? Ignore. */ 1168 tmpl = (struct nf_conn *)skb->nfct; 1169 if (!nf_ct_is_template(tmpl)) { 1170 NF_CT_STAT_INC_ATOMIC(net, ignore); 1171 return NF_ACCEPT; 1172 } 1173 skb->nfct = NULL; 1174 } 1175 1176 /* rcu_read_lock()ed by nf_hook_slow */ 1177 l3proto = __nf_ct_l3proto_find(pf); 1178 ret = l3proto->get_l4proto(skb, skb_network_offset(skb), 1179 &dataoff, &protonum); 1180 if (ret <= 0) { 1181 pr_debug("not prepared to track yet or error occurred\n"); 1182 NF_CT_STAT_INC_ATOMIC(net, error); 1183 NF_CT_STAT_INC_ATOMIC(net, invalid); 1184 ret = -ret; 1185 goto out; 1186 } 1187 1188 l4proto = __nf_ct_l4proto_find(pf, protonum); 1189 1190 /* It may be an special packet, error, unclean... 1191 * inverse of the return code tells to the netfilter 1192 * core what to do with the packet. */ 1193 if (l4proto->error != NULL) { 1194 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo, 1195 pf, hooknum); 1196 if (ret <= 0) { 1197 NF_CT_STAT_INC_ATOMIC(net, error); 1198 NF_CT_STAT_INC_ATOMIC(net, invalid); 1199 ret = -ret; 1200 goto out; 1201 } 1202 /* ICMP[v6] protocol trackers may assign one conntrack. */ 1203 if (skb->nfct) 1204 goto out; 1205 } 1206 1207 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, 1208 l3proto, l4proto, &set_reply, &ctinfo); 1209 if (!ct) { 1210 /* Not valid part of a connection */ 1211 NF_CT_STAT_INC_ATOMIC(net, invalid); 1212 ret = NF_ACCEPT; 1213 goto out; 1214 } 1215 1216 if (IS_ERR(ct)) { 1217 /* Too stressed to deal. */ 1218 NF_CT_STAT_INC_ATOMIC(net, drop); 1219 ret = NF_DROP; 1220 goto out; 1221 } 1222 1223 NF_CT_ASSERT(skb->nfct); 1224 1225 /* Decide what timeout policy we want to apply to this flow. */ 1226 timeouts = nf_ct_timeout_lookup(net, ct, l4proto); 1227 1228 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts); 1229 if (ret <= 0) { 1230 /* Invalid: inverse of the return code tells 1231 * the netfilter core what to do */ 1232 pr_debug("nf_conntrack_in: Can't track with proto module\n"); 1233 nf_conntrack_put(skb->nfct); 1234 skb->nfct = NULL; 1235 NF_CT_STAT_INC_ATOMIC(net, invalid); 1236 if (ret == -NF_DROP) 1237 NF_CT_STAT_INC_ATOMIC(net, drop); 1238 ret = -ret; 1239 goto out; 1240 } 1241 1242 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) 1243 nf_conntrack_event_cache(IPCT_REPLY, ct); 1244 out: 1245 if (tmpl) { 1246 /* Special case: we have to repeat this hook, assign the 1247 * template again to this packet. We assume that this packet 1248 * has no conntrack assigned. This is used by nf_ct_tcp. */ 1249 if (ret == NF_REPEAT) 1250 skb->nfct = (struct nf_conntrack *)tmpl; 1251 else 1252 nf_ct_put(tmpl); 1253 } 1254 1255 return ret; 1256 } 1257 EXPORT_SYMBOL_GPL(nf_conntrack_in); 1258 1259 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, 1260 const struct nf_conntrack_tuple *orig) 1261 { 1262 bool ret; 1263 1264 rcu_read_lock(); 1265 ret = nf_ct_invert_tuple(inverse, orig, 1266 __nf_ct_l3proto_find(orig->src.l3num), 1267 __nf_ct_l4proto_find(orig->src.l3num, 1268 orig->dst.protonum)); 1269 rcu_read_unlock(); 1270 return ret; 1271 } 1272 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); 1273 1274 /* Alter reply tuple (maybe alter helper). This is for NAT, and is 1275 implicitly racy: see __nf_conntrack_confirm */ 1276 void nf_conntrack_alter_reply(struct nf_conn *ct, 1277 const struct nf_conntrack_tuple *newreply) 1278 { 1279 struct nf_conn_help *help = nfct_help(ct); 1280 1281 /* Should be unconfirmed, so not in hash table yet */ 1282 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 1283 1284 pr_debug("Altering reply tuple of %p to ", ct); 1285 nf_ct_dump_tuple(newreply); 1286 1287 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; 1288 if (ct->master || (help && !hlist_empty(&help->expectations))) 1289 return; 1290 1291 rcu_read_lock(); 1292 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); 1293 rcu_read_unlock(); 1294 } 1295 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); 1296 1297 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ 1298 void __nf_ct_refresh_acct(struct nf_conn *ct, 1299 enum ip_conntrack_info ctinfo, 1300 const struct sk_buff *skb, 1301 unsigned long extra_jiffies, 1302 int do_acct) 1303 { 1304 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); 1305 NF_CT_ASSERT(skb); 1306 1307 /* Only update if this is not a fixed timeout */ 1308 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) 1309 goto acct; 1310 1311 /* If not in hash table, timer will not be active yet */ 1312 if (!nf_ct_is_confirmed(ct)) { 1313 ct->timeout.expires = extra_jiffies; 1314 } else { 1315 unsigned long newtime = jiffies + extra_jiffies; 1316 1317 /* Only update the timeout if the new timeout is at least 1318 HZ jiffies from the old timeout. Need del_timer for race 1319 avoidance (may already be dying). */ 1320 if (newtime - ct->timeout.expires >= HZ) 1321 mod_timer_pending(&ct->timeout, newtime); 1322 } 1323 1324 acct: 1325 if (do_acct) 1326 nf_ct_acct_update(ct, ctinfo, skb->len); 1327 } 1328 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); 1329 1330 bool __nf_ct_kill_acct(struct nf_conn *ct, 1331 enum ip_conntrack_info ctinfo, 1332 const struct sk_buff *skb, 1333 int do_acct) 1334 { 1335 if (do_acct) 1336 nf_ct_acct_update(ct, ctinfo, skb->len); 1337 1338 if (del_timer(&ct->timeout)) { 1339 ct->timeout.function((unsigned long)ct); 1340 return true; 1341 } 1342 return false; 1343 } 1344 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); 1345 1346 #ifdef CONFIG_NF_CONNTRACK_ZONES 1347 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = { 1348 .len = sizeof(struct nf_conntrack_zone), 1349 .align = __alignof__(struct nf_conntrack_zone), 1350 .id = NF_CT_EXT_ZONE, 1351 }; 1352 #endif 1353 1354 #if IS_ENABLED(CONFIG_NF_CT_NETLINK) 1355 1356 #include <linux/netfilter/nfnetlink.h> 1357 #include <linux/netfilter/nfnetlink_conntrack.h> 1358 #include <linux/mutex.h> 1359 1360 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be 1361 * in ip_conntrack_core, since we don't want the protocols to autoload 1362 * or depend on ctnetlink */ 1363 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, 1364 const struct nf_conntrack_tuple *tuple) 1365 { 1366 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) || 1367 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port)) 1368 goto nla_put_failure; 1369 return 0; 1370 1371 nla_put_failure: 1372 return -1; 1373 } 1374 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); 1375 1376 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { 1377 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, 1378 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, 1379 }; 1380 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); 1381 1382 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], 1383 struct nf_conntrack_tuple *t) 1384 { 1385 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) 1386 return -EINVAL; 1387 1388 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); 1389 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); 1390 1391 return 0; 1392 } 1393 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); 1394 1395 int nf_ct_port_nlattr_tuple_size(void) 1396 { 1397 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); 1398 } 1399 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); 1400 #endif 1401 1402 /* Used by ipt_REJECT and ip6t_REJECT. */ 1403 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb) 1404 { 1405 struct nf_conn *ct; 1406 enum ip_conntrack_info ctinfo; 1407 1408 /* This ICMP is in reverse direction to the packet which caused it */ 1409 ct = nf_ct_get(skb, &ctinfo); 1410 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) 1411 ctinfo = IP_CT_RELATED_REPLY; 1412 else 1413 ctinfo = IP_CT_RELATED; 1414 1415 /* Attach to new skbuff, and increment count */ 1416 nskb->nfct = &ct->ct_general; 1417 nskb->nfctinfo = ctinfo; 1418 nf_conntrack_get(nskb->nfct); 1419 } 1420 1421 /* Bring out ya dead! */ 1422 static struct nf_conn * 1423 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), 1424 void *data, unsigned int *bucket) 1425 { 1426 struct nf_conntrack_tuple_hash *h; 1427 struct nf_conn *ct; 1428 struct hlist_nulls_node *n; 1429 int cpu; 1430 spinlock_t *lockp; 1431 1432 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { 1433 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS]; 1434 local_bh_disable(); 1435 nf_conntrack_lock(lockp); 1436 if (*bucket < nf_conntrack_htable_size) { 1437 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) { 1438 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL) 1439 continue; 1440 ct = nf_ct_tuplehash_to_ctrack(h); 1441 if (net_eq(nf_ct_net(ct), net) && 1442 iter(ct, data)) 1443 goto found; 1444 } 1445 } 1446 spin_unlock(lockp); 1447 local_bh_enable(); 1448 cond_resched(); 1449 } 1450 1451 for_each_possible_cpu(cpu) { 1452 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); 1453 1454 spin_lock_bh(&pcpu->lock); 1455 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) { 1456 ct = nf_ct_tuplehash_to_ctrack(h); 1457 if (iter(ct, data)) 1458 set_bit(IPS_DYING_BIT, &ct->status); 1459 } 1460 spin_unlock_bh(&pcpu->lock); 1461 cond_resched(); 1462 } 1463 return NULL; 1464 found: 1465 atomic_inc(&ct->ct_general.use); 1466 spin_unlock(lockp); 1467 local_bh_enable(); 1468 return ct; 1469 } 1470 1471 void nf_ct_iterate_cleanup(struct net *net, 1472 int (*iter)(struct nf_conn *i, void *data), 1473 void *data, u32 portid, int report) 1474 { 1475 struct nf_conn *ct; 1476 unsigned int bucket = 0; 1477 1478 might_sleep(); 1479 1480 if (atomic_read(&net->ct.count) == 0) 1481 return; 1482 1483 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { 1484 /* Time to push up daises... */ 1485 if (del_timer(&ct->timeout)) 1486 nf_ct_delete(ct, portid, report); 1487 1488 /* ... else the timer will get him soon. */ 1489 1490 nf_ct_put(ct); 1491 cond_resched(); 1492 } 1493 } 1494 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); 1495 1496 static int kill_all(struct nf_conn *i, void *data) 1497 { 1498 return 1; 1499 } 1500 1501 void nf_ct_free_hashtable(void *hash, unsigned int size) 1502 { 1503 if (is_vmalloc_addr(hash)) 1504 vfree(hash); 1505 else 1506 free_pages((unsigned long)hash, 1507 get_order(sizeof(struct hlist_head) * size)); 1508 } 1509 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); 1510 1511 static int untrack_refs(void) 1512 { 1513 int cnt = 0, cpu; 1514 1515 for_each_possible_cpu(cpu) { 1516 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); 1517 1518 cnt += atomic_read(&ct->ct_general.use) - 1; 1519 } 1520 return cnt; 1521 } 1522 1523 void nf_conntrack_cleanup_start(void) 1524 { 1525 RCU_INIT_POINTER(ip_ct_attach, NULL); 1526 } 1527 1528 void nf_conntrack_cleanup_end(void) 1529 { 1530 RCU_INIT_POINTER(nf_ct_destroy, NULL); 1531 while (untrack_refs() > 0) 1532 schedule(); 1533 1534 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); 1535 1536 #ifdef CONFIG_NF_CONNTRACK_ZONES 1537 nf_ct_extend_unregister(&nf_ct_zone_extend); 1538 #endif 1539 nf_conntrack_proto_fini(); 1540 nf_conntrack_seqadj_fini(); 1541 nf_conntrack_labels_fini(); 1542 nf_conntrack_helper_fini(); 1543 nf_conntrack_timeout_fini(); 1544 nf_conntrack_ecache_fini(); 1545 nf_conntrack_tstamp_fini(); 1546 nf_conntrack_acct_fini(); 1547 nf_conntrack_expect_fini(); 1548 1549 kmem_cache_destroy(nf_conntrack_cachep); 1550 } 1551 1552 /* 1553 * Mishearing the voices in his head, our hero wonders how he's 1554 * supposed to kill the mall. 1555 */ 1556 void nf_conntrack_cleanup_net(struct net *net) 1557 { 1558 LIST_HEAD(single); 1559 1560 list_add(&net->exit_list, &single); 1561 nf_conntrack_cleanup_net_list(&single); 1562 } 1563 1564 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list) 1565 { 1566 int busy; 1567 struct net *net; 1568 1569 /* 1570 * This makes sure all current packets have passed through 1571 * netfilter framework. Roll on, two-stage module 1572 * delete... 1573 */ 1574 synchronize_net(); 1575 i_see_dead_people: 1576 busy = 0; 1577 list_for_each_entry(net, net_exit_list, exit_list) { 1578 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0); 1579 if (atomic_read(&net->ct.count) != 0) 1580 busy = 1; 1581 } 1582 if (busy) { 1583 schedule(); 1584 goto i_see_dead_people; 1585 } 1586 1587 list_for_each_entry(net, net_exit_list, exit_list) { 1588 nf_conntrack_proto_pernet_fini(net); 1589 nf_conntrack_helper_pernet_fini(net); 1590 nf_conntrack_ecache_pernet_fini(net); 1591 nf_conntrack_tstamp_pernet_fini(net); 1592 nf_conntrack_acct_pernet_fini(net); 1593 nf_conntrack_expect_pernet_fini(net); 1594 free_percpu(net->ct.stat); 1595 free_percpu(net->ct.pcpu_lists); 1596 } 1597 } 1598 1599 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls) 1600 { 1601 struct hlist_nulls_head *hash; 1602 unsigned int nr_slots, i; 1603 size_t sz; 1604 1605 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head))) 1606 return NULL; 1607 1608 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); 1609 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); 1610 1611 if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head))) 1612 return NULL; 1613 1614 sz = nr_slots * sizeof(struct hlist_nulls_head); 1615 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, 1616 get_order(sz)); 1617 if (!hash) 1618 hash = vzalloc(sz); 1619 1620 if (hash && nulls) 1621 for (i = 0; i < nr_slots; i++) 1622 INIT_HLIST_NULLS_HEAD(&hash[i], i); 1623 1624 return hash; 1625 } 1626 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); 1627 1628 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) 1629 { 1630 int i, bucket, rc; 1631 unsigned int hashsize, old_size; 1632 struct hlist_nulls_head *hash, *old_hash; 1633 struct nf_conntrack_tuple_hash *h; 1634 struct nf_conn *ct; 1635 1636 if (current->nsproxy->net_ns != &init_net) 1637 return -EOPNOTSUPP; 1638 1639 /* On boot, we can set this without any fancy locking. */ 1640 if (!nf_conntrack_htable_size) 1641 return param_set_uint(val, kp); 1642 1643 rc = kstrtouint(val, 0, &hashsize); 1644 if (rc) 1645 return rc; 1646 if (!hashsize) 1647 return -EINVAL; 1648 1649 hash = nf_ct_alloc_hashtable(&hashsize, 1); 1650 if (!hash) 1651 return -ENOMEM; 1652 1653 local_bh_disable(); 1654 nf_conntrack_all_lock(); 1655 write_seqcount_begin(&nf_conntrack_generation); 1656 1657 /* Lookups in the old hash might happen in parallel, which means we 1658 * might get false negatives during connection lookup. New connections 1659 * created because of a false negative won't make it into the hash 1660 * though since that required taking the locks. 1661 */ 1662 1663 for (i = 0; i < nf_conntrack_htable_size; i++) { 1664 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) { 1665 h = hlist_nulls_entry(nf_conntrack_hash[i].first, 1666 struct nf_conntrack_tuple_hash, hnnode); 1667 ct = nf_ct_tuplehash_to_ctrack(h); 1668 hlist_nulls_del_rcu(&h->hnnode); 1669 bucket = __hash_conntrack(nf_ct_net(ct), 1670 &h->tuple, hashsize); 1671 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); 1672 } 1673 } 1674 old_size = nf_conntrack_htable_size; 1675 old_hash = nf_conntrack_hash; 1676 1677 nf_conntrack_hash = hash; 1678 nf_conntrack_htable_size = hashsize; 1679 1680 write_seqcount_end(&nf_conntrack_generation); 1681 nf_conntrack_all_unlock(); 1682 local_bh_enable(); 1683 1684 synchronize_net(); 1685 nf_ct_free_hashtable(old_hash, old_size); 1686 return 0; 1687 } 1688 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); 1689 1690 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, 1691 &nf_conntrack_htable_size, 0600); 1692 1693 void nf_ct_untracked_status_or(unsigned long bits) 1694 { 1695 int cpu; 1696 1697 for_each_possible_cpu(cpu) 1698 per_cpu(nf_conntrack_untracked, cpu).status |= bits; 1699 } 1700 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or); 1701 1702 int nf_conntrack_init_start(void) 1703 { 1704 int max_factor = 8; 1705 int ret = -ENOMEM; 1706 int i, cpu; 1707 1708 seqcount_init(&nf_conntrack_generation); 1709 1710 for (i = 0; i < CONNTRACK_LOCKS; i++) 1711 spin_lock_init(&nf_conntrack_locks[i]); 1712 1713 if (!nf_conntrack_htable_size) { 1714 /* Idea from tcp.c: use 1/16384 of memory. 1715 * On i386: 32MB machine has 512 buckets. 1716 * >= 1GB machines have 16384 buckets. 1717 * >= 4GB machines have 65536 buckets. 1718 */ 1719 nf_conntrack_htable_size 1720 = (((totalram_pages << PAGE_SHIFT) / 16384) 1721 / sizeof(struct hlist_head)); 1722 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE))) 1723 nf_conntrack_htable_size = 65536; 1724 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) 1725 nf_conntrack_htable_size = 16384; 1726 if (nf_conntrack_htable_size < 32) 1727 nf_conntrack_htable_size = 32; 1728 1729 /* Use a max. factor of four by default to get the same max as 1730 * with the old struct list_heads. When a table size is given 1731 * we use the old value of 8 to avoid reducing the max. 1732 * entries. */ 1733 max_factor = 4; 1734 } 1735 1736 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1); 1737 if (!nf_conntrack_hash) 1738 return -ENOMEM; 1739 1740 nf_conntrack_max = max_factor * nf_conntrack_htable_size; 1741 1742 nf_conntrack_cachep = kmem_cache_create("nf_conntrack", 1743 sizeof(struct nf_conn), 0, 1744 SLAB_DESTROY_BY_RCU, NULL); 1745 if (!nf_conntrack_cachep) 1746 goto err_cachep; 1747 1748 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", 1749 NF_CONNTRACK_VERSION, nf_conntrack_htable_size, 1750 nf_conntrack_max); 1751 1752 ret = nf_conntrack_expect_init(); 1753 if (ret < 0) 1754 goto err_expect; 1755 1756 ret = nf_conntrack_acct_init(); 1757 if (ret < 0) 1758 goto err_acct; 1759 1760 ret = nf_conntrack_tstamp_init(); 1761 if (ret < 0) 1762 goto err_tstamp; 1763 1764 ret = nf_conntrack_ecache_init(); 1765 if (ret < 0) 1766 goto err_ecache; 1767 1768 ret = nf_conntrack_timeout_init(); 1769 if (ret < 0) 1770 goto err_timeout; 1771 1772 ret = nf_conntrack_helper_init(); 1773 if (ret < 0) 1774 goto err_helper; 1775 1776 ret = nf_conntrack_labels_init(); 1777 if (ret < 0) 1778 goto err_labels; 1779 1780 ret = nf_conntrack_seqadj_init(); 1781 if (ret < 0) 1782 goto err_seqadj; 1783 1784 #ifdef CONFIG_NF_CONNTRACK_ZONES 1785 ret = nf_ct_extend_register(&nf_ct_zone_extend); 1786 if (ret < 0) 1787 goto err_extend; 1788 #endif 1789 ret = nf_conntrack_proto_init(); 1790 if (ret < 0) 1791 goto err_proto; 1792 1793 /* Set up fake conntrack: to never be deleted, not in any hashes */ 1794 for_each_possible_cpu(cpu) { 1795 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); 1796 write_pnet(&ct->ct_net, &init_net); 1797 atomic_set(&ct->ct_general.use, 1); 1798 } 1799 /* - and look it like as a confirmed connection */ 1800 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED); 1801 return 0; 1802 1803 err_proto: 1804 #ifdef CONFIG_NF_CONNTRACK_ZONES 1805 nf_ct_extend_unregister(&nf_ct_zone_extend); 1806 err_extend: 1807 #endif 1808 nf_conntrack_seqadj_fini(); 1809 err_seqadj: 1810 nf_conntrack_labels_fini(); 1811 err_labels: 1812 nf_conntrack_helper_fini(); 1813 err_helper: 1814 nf_conntrack_timeout_fini(); 1815 err_timeout: 1816 nf_conntrack_ecache_fini(); 1817 err_ecache: 1818 nf_conntrack_tstamp_fini(); 1819 err_tstamp: 1820 nf_conntrack_acct_fini(); 1821 err_acct: 1822 nf_conntrack_expect_fini(); 1823 err_expect: 1824 kmem_cache_destroy(nf_conntrack_cachep); 1825 err_cachep: 1826 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size); 1827 return ret; 1828 } 1829 1830 void nf_conntrack_init_end(void) 1831 { 1832 /* For use by REJECT target */ 1833 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach); 1834 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack); 1835 } 1836 1837 /* 1838 * We need to use special "null" values, not used in hash table 1839 */ 1840 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0) 1841 #define DYING_NULLS_VAL ((1<<30)+1) 1842 #define TEMPLATE_NULLS_VAL ((1<<30)+2) 1843 1844 int nf_conntrack_init_net(struct net *net) 1845 { 1846 int ret = -ENOMEM; 1847 int cpu; 1848 1849 atomic_set(&net->ct.count, 0); 1850 1851 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu); 1852 if (!net->ct.pcpu_lists) 1853 goto err_stat; 1854 1855 for_each_possible_cpu(cpu) { 1856 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu); 1857 1858 spin_lock_init(&pcpu->lock); 1859 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL); 1860 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL); 1861 } 1862 1863 net->ct.stat = alloc_percpu(struct ip_conntrack_stat); 1864 if (!net->ct.stat) 1865 goto err_pcpu_lists; 1866 1867 ret = nf_conntrack_expect_pernet_init(net); 1868 if (ret < 0) 1869 goto err_expect; 1870 ret = nf_conntrack_acct_pernet_init(net); 1871 if (ret < 0) 1872 goto err_acct; 1873 ret = nf_conntrack_tstamp_pernet_init(net); 1874 if (ret < 0) 1875 goto err_tstamp; 1876 ret = nf_conntrack_ecache_pernet_init(net); 1877 if (ret < 0) 1878 goto err_ecache; 1879 ret = nf_conntrack_helper_pernet_init(net); 1880 if (ret < 0) 1881 goto err_helper; 1882 ret = nf_conntrack_proto_pernet_init(net); 1883 if (ret < 0) 1884 goto err_proto; 1885 return 0; 1886 1887 err_proto: 1888 nf_conntrack_helper_pernet_fini(net); 1889 err_helper: 1890 nf_conntrack_ecache_pernet_fini(net); 1891 err_ecache: 1892 nf_conntrack_tstamp_pernet_fini(net); 1893 err_tstamp: 1894 nf_conntrack_acct_pernet_fini(net); 1895 err_acct: 1896 nf_conntrack_expect_pernet_fini(net); 1897 err_expect: 1898 free_percpu(net->ct.stat); 1899 err_pcpu_lists: 1900 free_percpu(net->ct.pcpu_lists); 1901 err_stat: 1902 return ret; 1903 } 1904