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  * (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_l4proto.h>
  41 #include <net/netfilter/nf_conntrack_expect.h>
  42 #include <net/netfilter/nf_conntrack_helper.h>
  43 #include <net/netfilter/nf_conntrack_seqadj.h>
  44 #include <net/netfilter/nf_conntrack_core.h>
  45 #include <net/netfilter/nf_conntrack_extend.h>
  46 #include <net/netfilter/nf_conntrack_acct.h>
  47 #include <net/netfilter/nf_conntrack_ecache.h>
  48 #include <net/netfilter/nf_conntrack_zones.h>
  49 #include <net/netfilter/nf_conntrack_timestamp.h>
  50 #include <net/netfilter/nf_conntrack_timeout.h>
  51 #include <net/netfilter/nf_conntrack_labels.h>
  52 #include <net/netfilter/nf_conntrack_synproxy.h>
  53 #include <net/netfilter/nf_nat.h>
  54 #include <net/netfilter/nf_nat_core.h>
  55 #include <net/netfilter/nf_nat_helper.h>
  56 #include <net/netns/hash.h>
  57 #include <net/ip.h>
  58
  59 #include "nf_internals.h"
  60
  61 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
  62 EXPORT_SYMBOL_GPL(nf_conntrack_locks);
  63
  64 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
  65 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
  66
  67 struct hlist_nulls_head *nf_conntrack_hash __read_mostly;
  68 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
  69
  70 struct conntrack_gc_work {
  71         struct delayed_work     dwork;
  72         u32                     last_bucket;
  73         bool                    exiting;
  74         bool                    early_drop;
  75         long                    next_gc_run;
  76 };
  77
  78 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
  79 static __read_mostly spinlock_t nf_conntrack_locks_all_lock;
  80 static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
  81 static __read_mostly bool nf_conntrack_locks_all;
  82
  83 /* every gc cycle scans at most 1/GC_MAX_BUCKETS_DIV part of table */
  84 #define GC_MAX_BUCKETS_DIV      128u
  85 /* upper bound of full table scan */
  86 #define GC_MAX_SCAN_JIFFIES     (16u * HZ)
  87 /* desired ratio of entries found to be expired */
  88 #define GC_EVICT_RATIO  50u
  89
  90 static struct conntrack_gc_work conntrack_gc_work;
  91
  92 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
  93 {
  94         /* 1) Acquire the lock */
  95         spin_lock(lock);
  96
  97         /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
  98          * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
  99          */
 100         if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false))
 101                 return;
 102
 103         /* fast path failed, unlock */
 104         spin_unlock(lock);
 105
 106         /* Slow path 1) get global lock */
 107         spin_lock(&nf_conntrack_locks_all_lock);
 108
 109         /* Slow path 2) get the lock we want */
 110         spin_lock(lock);
 111
 112         /* Slow path 3) release the global lock */
 113         spin_unlock(&nf_conntrack_locks_all_lock);
 114 }
 115 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
 116
 117 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
 118 {
 119         h1 %= CONNTRACK_LOCKS;
 120         h2 %= CONNTRACK_LOCKS;
 121         spin_unlock(&nf_conntrack_locks[h1]);
 122         if (h1 != h2)
 123                 spin_unlock(&nf_conntrack_locks[h2]);
 124 }
 125
 126 /* return true if we need to recompute hashes (in case hash table was resized) */
 127 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
 128                                      unsigned int h2, unsigned int sequence)
 129 {
 130         h1 %= CONNTRACK_LOCKS;
 131         h2 %= CONNTRACK_LOCKS;
 132         if (h1 <= h2) {
 133                 nf_conntrack_lock(&nf_conntrack_locks[h1]);
 134                 if (h1 != h2)
 135                         spin_lock_nested(&nf_conntrack_locks[h2],
 136                                          SINGLE_DEPTH_NESTING);
 137         } else {
 138                 nf_conntrack_lock(&nf_conntrack_locks[h2]);
 139                 spin_lock_nested(&nf_conntrack_locks[h1],
 140                                  SINGLE_DEPTH_NESTING);
 141         }
 142         if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
 143                 nf_conntrack_double_unlock(h1, h2);
 144                 return true;
 145         }
 146         return false;
 147 }
 148
 149 static void nf_conntrack_all_lock(void)
 150 {
 151         int i;
 152
 153         spin_lock(&nf_conntrack_locks_all_lock);
 154
 155         nf_conntrack_locks_all = true;
 156
 157         for (i = 0; i < CONNTRACK_LOCKS; i++) {
 158                 spin_lock(&nf_conntrack_locks[i]);
 159
 160                 /* This spin_unlock provides the "release" to ensure that
 161                  * nf_conntrack_locks_all==true is visible to everyone that
 162                  * acquired spin_lock(&nf_conntrack_locks[]).
 163                  */
 164                 spin_unlock(&nf_conntrack_locks[i]);
 165         }
 166 }
 167
 168 static void nf_conntrack_all_unlock(void)
 169 {
 170         /* All prior stores must be complete before we clear
 171          * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
 172          * might observe the false value but not the entire
 173          * critical section.
 174          * It pairs with the smp_load_acquire() in nf_conntrack_lock()
 175          */
 176         smp_store_release(&nf_conntrack_locks_all, false);
 177         spin_unlock(&nf_conntrack_locks_all_lock);
 178 }
 179
 180 unsigned int nf_conntrack_htable_size __read_mostly;
 181 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
 182
 183 unsigned int nf_conntrack_max __read_mostly;
 184 EXPORT_SYMBOL_GPL(nf_conntrack_max);
 185 seqcount_t nf_conntrack_generation __read_mostly;
 186 static unsigned int nf_conntrack_hash_rnd __read_mostly;
 187
 188 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
 189                               const struct net *net)
 190 {
 191         unsigned int n;
 192         u32 seed;
 193
 194         get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
 195
 196         /* The direction must be ignored, so we hash everything up to the
 197          * destination ports (which is a multiple of 4) and treat the last
 198          * three bytes manually.
 199          */
 200         seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
 201         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
 202         return jhash2((u32 *)tuple, n, seed ^
 203                       (((__force __u16)tuple->dst.u.all << 16) |
 204                       tuple->dst.protonum));
 205 }
 206
 207 static u32 scale_hash(u32 hash)
 208 {
 209         return reciprocal_scale(hash, nf_conntrack_htable_size);
 210 }
 211
 212 static u32 __hash_conntrack(const struct net *net,
 213                             const struct nf_conntrack_tuple *tuple,
 214                             unsigned int size)
 215 {
 216         return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
 217 }
 218
 219 static u32 hash_conntrack(const struct net *net,
 220                           const struct nf_conntrack_tuple *tuple)
 221 {
 222         return scale_hash(hash_conntrack_raw(tuple, net));
 223 }
 224
 225 static bool
 226 nf_ct_get_tuple(const struct sk_buff *skb,
 227                 unsigned int nhoff,
 228                 unsigned int dataoff,
 229                 u_int16_t l3num,
 230                 u_int8_t protonum,
 231                 struct net *net,
 232                 struct nf_conntrack_tuple *tuple,
 233                 const struct nf_conntrack_l4proto *l4proto)
 234 {
 235         unsigned int size;
 236         const __be32 *ap;
 237         __be32 _addrs[8];
 238         struct {
 239                 __be16 sport;
 240                 __be16 dport;
 241         } _inet_hdr, *inet_hdr;
 242
 243         memset(tuple, 0, sizeof(*tuple));
 244
 245         tuple->src.l3num = l3num;
 246         switch (l3num) {
 247         case NFPROTO_IPV4:
 248                 nhoff += offsetof(struct iphdr, saddr);
 249                 size = 2 * sizeof(__be32);
 250                 break;
 251         case NFPROTO_IPV6:
 252                 nhoff += offsetof(struct ipv6hdr, saddr);
 253                 size = sizeof(_addrs);
 254                 break;
 255         default:
 256                 return true;
 257         }
 258
 259         ap = skb_header_pointer(skb, nhoff, size, _addrs);
 260         if (!ap)
 261                 return false;
 262
 263         switch (l3num) {
 264         case NFPROTO_IPV4:
 265                 tuple->src.u3.ip = ap[0];
 266                 tuple->dst.u3.ip = ap[1];
 267                 break;
 268         case NFPROTO_IPV6:
 269                 memcpy(tuple->src.u3.ip6, ap, sizeof(tuple->src.u3.ip6));
 270                 memcpy(tuple->dst.u3.ip6, ap + 4, sizeof(tuple->dst.u3.ip6));
 271                 break;
 272         }
 273
 274         tuple->dst.protonum = protonum;
 275         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 276
 277         if (unlikely(l4proto->pkt_to_tuple))
 278                 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
 279
 280         /* Actually only need first 4 bytes to get ports. */
 281         inet_hdr = skb_header_pointer(skb, dataoff, sizeof(_inet_hdr), &_inet_hdr);
 282         if (!inet_hdr)
 283                 return false;
 284
 285         tuple->src.u.udp.port = inet_hdr->sport;
 286         tuple->dst.u.udp.port = inet_hdr->dport;
 287         return true;
 288 }
 289
 290 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
 291                             u_int8_t *protonum)
 292 {
 293         int dataoff = -1;
 294         const struct iphdr *iph;
 295         struct iphdr _iph;
 296
 297         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
 298         if (!iph)
 299                 return -1;
 300
 301         /* Conntrack defragments packets, we might still see fragments
 302          * inside ICMP packets though.
 303          */
 304         if (iph->frag_off & htons(IP_OFFSET))
 305                 return -1;
 306
 307         dataoff = nhoff + (iph->ihl << 2);
 308         *protonum = iph->protocol;
 309
 310         /* Check bogus IP headers */
 311         if (dataoff > skb->len) {
 312                 pr_debug("bogus IPv4 packet: nhoff %u, ihl %u, skblen %u\n",
 313                          nhoff, iph->ihl << 2, skb->len);
 314                 return -1;
 315         }
 316         return dataoff;
 317 }
 318
 319 #if IS_ENABLED(CONFIG_IPV6)
 320 static int ipv6_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
 321                             u8 *protonum)
 322 {
 323         int protoff = -1;
 324         unsigned int extoff = nhoff + sizeof(struct ipv6hdr);
 325         __be16 frag_off;
 326         u8 nexthdr;
 327
 328         if (skb_copy_bits(skb, nhoff + offsetof(struct ipv6hdr, nexthdr),
 329                           &nexthdr, sizeof(nexthdr)) != 0) {
 330                 pr_debug("can't get nexthdr\n");
 331                 return -1;
 332         }
 333         protoff = ipv6_skip_exthdr(skb, extoff, &nexthdr, &frag_off);
 334         /*
 335          * (protoff == skb->len) means the packet has not data, just
 336          * IPv6 and possibly extensions headers, but it is tracked anyway
 337          */
 338         if (protoff < 0 || (frag_off & htons(~0x7)) != 0) {
 339                 pr_debug("can't find proto in pkt\n");
 340                 return -1;
 341         }
 342
 343         *protonum = nexthdr;
 344         return protoff;
 345 }
 346 #endif
 347
 348 static int get_l4proto(const struct sk_buff *skb,
 349                        unsigned int nhoff, u8 pf, u8 *l4num)
 350 {
 351         switch (pf) {
 352         case NFPROTO_IPV4:
 353                 return ipv4_get_l4proto(skb, nhoff, l4num);
 354 #if IS_ENABLED(CONFIG_IPV6)
 355         case NFPROTO_IPV6:
 356                 return ipv6_get_l4proto(skb, nhoff, l4num);
 357 #endif
 358         default:
 359                 *l4num = 0;
 360                 break;
 361         }
 362         return -1;
 363 }
 364
 365 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 366                        u_int16_t l3num,
 367                        struct net *net, struct nf_conntrack_tuple *tuple)
 368 {
 369         const struct nf_conntrack_l4proto *l4proto;
 370         u8 protonum;
 371         int protoff;
 372         int ret;
 373
 374         rcu_read_lock();
 375
 376         protoff = get_l4proto(skb, nhoff, l3num, &protonum);
 377         if (protoff <= 0) {
 378                 rcu_read_unlock();
 379                 return false;
 380         }
 381
 382         l4proto = __nf_ct_l4proto_find(protonum);
 383
 384         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
 385                               l4proto);
 386
 387         rcu_read_unlock();
 388         return ret;
 389 }
 390 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 391
 392 bool
 393 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 394                    const struct nf_conntrack_tuple *orig,
 395                    const struct nf_conntrack_l4proto *l4proto)
 396 {
 397         memset(inverse, 0, sizeof(*inverse));
 398
 399         inverse->src.l3num = orig->src.l3num;
 400
 401         switch (orig->src.l3num) {
 402         case NFPROTO_IPV4:
 403                 inverse->src.u3.ip = orig->dst.u3.ip;
 404                 inverse->dst.u3.ip = orig->src.u3.ip;
 405                 break;
 406         case NFPROTO_IPV6:
 407                 inverse->src.u3.in6 = orig->dst.u3.in6;
 408                 inverse->dst.u3.in6 = orig->src.u3.in6;
 409                 break;
 410         default:
 411                 break;
 412         }
 413
 414         inverse->dst.dir = !orig->dst.dir;
 415
 416         inverse->dst.protonum = orig->dst.protonum;
 417
 418         if (unlikely(l4proto->invert_tuple))
 419                 return l4proto->invert_tuple(inverse, orig);
 420
 421         inverse->src.u.all = orig->dst.u.all;
 422         inverse->dst.u.all = orig->src.u.all;
 423         return true;
 424 }
 425 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 426
 427 static void
 428 clean_from_lists(struct nf_conn *ct)
 429 {
 430         pr_debug("clean_from_lists(%p)\n", ct);
 431         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 432         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 433
 434         /* Destroy all pending expectations */
 435         nf_ct_remove_expectations(ct);
 436 }
 437
 438 /* must be called with local_bh_disable */
 439 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
 440 {
 441         struct ct_pcpu *pcpu;
 442
 443         /* add this conntrack to the (per cpu) dying list */
 444         ct->cpu = smp_processor_id();
 445         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 446
 447         spin_lock(&pcpu->lock);
 448         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 449                              &pcpu->dying);
 450         spin_unlock(&pcpu->lock);
 451 }
 452
 453 /* must be called with local_bh_disable */
 454 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
 455 {
 456         struct ct_pcpu *pcpu;
 457
 458         /* add this conntrack to the (per cpu) unconfirmed list */
 459         ct->cpu = smp_processor_id();
 460         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 461
 462         spin_lock(&pcpu->lock);
 463         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 464                              &pcpu->unconfirmed);
 465         spin_unlock(&pcpu->lock);
 466 }
 467
 468 /* must be called with local_bh_disable */
 469 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
 470 {
 471         struct ct_pcpu *pcpu;
 472
 473         /* We overload first tuple to link into unconfirmed or dying list.*/
 474         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 475
 476         spin_lock(&pcpu->lock);
 477         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 478         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 479         spin_unlock(&pcpu->lock);
 480 }
 481
 482 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
 483
 484 /* Released via destroy_conntrack() */
 485 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
 486                                  const struct nf_conntrack_zone *zone,
 487                                  gfp_t flags)
 488 {
 489         struct nf_conn *tmpl, *p;
 490
 491         if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) {
 492                 tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags);
 493                 if (!tmpl)
 494                         return NULL;
 495
 496                 p = tmpl;
 497                 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
 498                 if (tmpl != p) {
 499                         tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
 500                         tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p;
 501                 }
 502         } else {
 503                 tmpl = kzalloc(sizeof(*tmpl), flags);
 504                 if (!tmpl)
 505                         return NULL;
 506         }
 507
 508         tmpl->status = IPS_TEMPLATE;
 509         write_pnet(&tmpl->ct_net, net);
 510         nf_ct_zone_add(tmpl, zone);
 511         atomic_set(&tmpl->ct_general.use, 0);
 512
 513         return tmpl;
 514 }
 515 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
 516
 517 void nf_ct_tmpl_free(struct nf_conn *tmpl)
 518 {
 519         nf_ct_ext_destroy(tmpl);
 520         nf_ct_ext_free(tmpl);
 521
 522         if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK)
 523                 kfree((char *)tmpl - tmpl->proto.tmpl_padto);
 524         else
 525                 kfree(tmpl);
 526 }
 527 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
 528
 529 static void
 530 destroy_conntrack(struct nf_conntrack *nfct)
 531 {
 532         struct nf_conn *ct = (struct nf_conn *)nfct;
 533         const struct nf_conntrack_l4proto *l4proto;
 534
 535         pr_debug("destroy_conntrack(%p)\n", ct);
 536         WARN_ON(atomic_read(&nfct->use) != 0);
 537
 538         if (unlikely(nf_ct_is_template(ct))) {
 539                 nf_ct_tmpl_free(ct);
 540                 return;
 541         }
 542         l4proto = __nf_ct_l4proto_find(nf_ct_protonum(ct));
 543         if (l4proto->destroy)
 544                 l4proto->destroy(ct);
 545
 546         local_bh_disable();
 547         /* Expectations will have been removed in clean_from_lists,
 548          * except TFTP can create an expectation on the first packet,
 549          * before connection is in the list, so we need to clean here,
 550          * too.
 551          */
 552         nf_ct_remove_expectations(ct);
 553
 554         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 555
 556         local_bh_enable();
 557
 558         if (ct->master)
 559                 nf_ct_put(ct->master);
 560
 561         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 562         nf_conntrack_free(ct);
 563 }
 564
 565 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 566 {
 567         struct net *net = nf_ct_net(ct);
 568         unsigned int hash, reply_hash;
 569         unsigned int sequence;
 570
 571         nf_ct_helper_destroy(ct);
 572
 573         local_bh_disable();
 574         do {
 575                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 576                 hash = hash_conntrack(net,
 577                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 578                 reply_hash = hash_conntrack(net,
 579                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 580         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 581
 582         clean_from_lists(ct);
 583         nf_conntrack_double_unlock(hash, reply_hash);
 584
 585         nf_ct_add_to_dying_list(ct);
 586
 587         local_bh_enable();
 588 }
 589
 590 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 591 {
 592         struct nf_conn_tstamp *tstamp;
 593
 594         if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
 595                 return false;
 596
 597         tstamp = nf_conn_tstamp_find(ct);
 598         if (tstamp && tstamp->stop == 0)
 599                 tstamp->stop = ktime_get_real_ns();
 600
 601         if (nf_conntrack_event_report(IPCT_DESTROY, ct,
 602                                     portid, report) < 0) {
 603                 /* destroy event was not delivered. nf_ct_put will
 604                  * be done by event cache worker on redelivery.
 605                  */
 606                 nf_ct_delete_from_lists(ct);
 607                 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
 608                 return false;
 609         }
 610
 611         nf_conntrack_ecache_work(nf_ct_net(ct));
 612         nf_ct_delete_from_lists(ct);
 613         nf_ct_put(ct);
 614         return true;
 615 }
 616 EXPORT_SYMBOL_GPL(nf_ct_delete);
 617
 618 static inline bool
 619 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
 620                 const struct nf_conntrack_tuple *tuple,
 621                 const struct nf_conntrack_zone *zone,
 622                 const struct net *net)
 623 {
 624         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 625
 626         /* A conntrack can be recreated with the equal tuple,
 627          * so we need to check that the conntrack is confirmed
 628          */
 629         return nf_ct_tuple_equal(tuple, &h->tuple) &&
 630                nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
 631                nf_ct_is_confirmed(ct) &&
 632                net_eq(net, nf_ct_net(ct));
 633 }
 634
 635 static inline bool
 636 nf_ct_match(const struct nf_conn *ct1, const struct nf_conn *ct2)
 637 {
 638         return nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 639                                  &ct2->tuplehash[IP_CT_DIR_ORIGINAL].tuple) &&
 640                nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_REPLY].tuple,
 641                                  &ct2->tuplehash[IP_CT_DIR_REPLY].tuple) &&
 642                nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_ORIGINAL) &&
 643                nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_REPLY) &&
 644                net_eq(nf_ct_net(ct1), nf_ct_net(ct2));
 645 }
 646
 647 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
 648 static void nf_ct_gc_expired(struct nf_conn *ct)
 649 {
 650         if (!atomic_inc_not_zero(&ct->ct_general.use))
 651                 return;
 652
 653         if (nf_ct_should_gc(ct))
 654                 nf_ct_kill(ct);
 655
 656         nf_ct_put(ct);
 657 }
 658
 659 /*
 660  * Warning :
 661  * - Caller must take a reference on returned object
 662  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 663  */
 664 static struct nf_conntrack_tuple_hash *
 665 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
 666                       const struct nf_conntrack_tuple *tuple, u32 hash)
 667 {
 668         struct nf_conntrack_tuple_hash *h;
 669         struct hlist_nulls_head *ct_hash;
 670         struct hlist_nulls_node *n;
 671         unsigned int bucket, hsize;
 672
 673 begin:
 674         nf_conntrack_get_ht(&ct_hash, &hsize);
 675         bucket = reciprocal_scale(hash, hsize);
 676
 677         hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
 678                 struct nf_conn *ct;
 679
 680                 ct = nf_ct_tuplehash_to_ctrack(h);
 681                 if (nf_ct_is_expired(ct)) {
 682                         nf_ct_gc_expired(ct);
 683                         continue;
 684                 }
 685
 686                 if (nf_ct_is_dying(ct))
 687                         continue;
 688
 689                 if (nf_ct_key_equal(h, tuple, zone, net))
 690                         return h;
 691         }
 692         /*
 693          * if the nulls value we got at the end of this lookup is
 694          * not the expected one, we must restart lookup.
 695          * We probably met an item that was moved to another chain.
 696          */
 697         if (get_nulls_value(n) != bucket) {
 698                 NF_CT_STAT_INC_ATOMIC(net, search_restart);
 699                 goto begin;
 700         }
 701
 702         return NULL;
 703 }
 704
 705 /* Find a connection corresponding to a tuple. */
 706 static struct nf_conntrack_tuple_hash *
 707 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 708                         const struct nf_conntrack_tuple *tuple, u32 hash)
 709 {
 710         struct nf_conntrack_tuple_hash *h;
 711         struct nf_conn *ct;
 712
 713         rcu_read_lock();
 714 begin:
 715         h = ____nf_conntrack_find(net, zone, tuple, hash);
 716         if (h) {
 717                 ct = nf_ct_tuplehash_to_ctrack(h);
 718                 if (unlikely(nf_ct_is_dying(ct) ||
 719                              !atomic_inc_not_zero(&ct->ct_general.use)))
 720                         h = NULL;
 721                 else {
 722                         if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) {
 723                                 nf_ct_put(ct);
 724                                 goto begin;
 725                         }
 726                 }
 727         }
 728         rcu_read_unlock();
 729
 730         return h;
 731 }
 732
 733 struct nf_conntrack_tuple_hash *
 734 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 735                       const struct nf_conntrack_tuple *tuple)
 736 {
 737         return __nf_conntrack_find_get(net, zone, tuple,
 738                                        hash_conntrack_raw(tuple, net));
 739 }
 740 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 741
 742 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 743                                        unsigned int hash,
 744                                        unsigned int reply_hash)
 745 {
 746         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 747                            &nf_conntrack_hash[hash]);
 748         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 749                            &nf_conntrack_hash[reply_hash]);
 750 }
 751
 752 int
 753 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 754 {
 755         const struct nf_conntrack_zone *zone;
 756         struct net *net = nf_ct_net(ct);
 757         unsigned int hash, reply_hash;
 758         struct nf_conntrack_tuple_hash *h;
 759         struct hlist_nulls_node *n;
 760         unsigned int sequence;
 761
 762         zone = nf_ct_zone(ct);
 763
 764         local_bh_disable();
 765         do {
 766                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 767                 hash = hash_conntrack(net,
 768                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 769                 reply_hash = hash_conntrack(net,
 770                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 771         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 772
 773         /* See if there's one in the list already, including reverse */
 774         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
 775                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 776                                     zone, net))
 777                         goto out;
 778
 779         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
 780                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 781                                     zone, net))
 782                         goto out;
 783
 784         smp_wmb();
 785         /* The caller holds a reference to this object */
 786         atomic_set(&ct->ct_general.use, 2);
 787         __nf_conntrack_hash_insert(ct, hash, reply_hash);
 788         nf_conntrack_double_unlock(hash, reply_hash);
 789         NF_CT_STAT_INC(net, insert);
 790         local_bh_enable();
 791         return 0;
 792
 793 out:
 794         nf_conntrack_double_unlock(hash, reply_hash);
 795         NF_CT_STAT_INC(net, insert_failed);
 796         local_bh_enable();
 797         return -EEXIST;
 798 }
 799 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 800
 801 static inline void nf_ct_acct_update(struct nf_conn *ct,
 802                                      enum ip_conntrack_info ctinfo,
 803                                      unsigned int len)
 804 {
 805         struct nf_conn_acct *acct;
 806
 807         acct = nf_conn_acct_find(ct);
 808         if (acct) {
 809                 struct nf_conn_counter *counter = acct->counter;
 810
 811                 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
 812                 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
 813         }
 814 }
 815
 816 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
 817                              const struct nf_conn *loser_ct)
 818 {
 819         struct nf_conn_acct *acct;
 820
 821         acct = nf_conn_acct_find(loser_ct);
 822         if (acct) {
 823                 struct nf_conn_counter *counter = acct->counter;
 824                 unsigned int bytes;
 825
 826                 /* u32 should be fine since we must have seen one packet. */
 827                 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
 828                 nf_ct_acct_update(ct, ctinfo, bytes);
 829         }
 830 }
 831
 832 /* Resolve race on insertion if this protocol allows this. */
 833 static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
 834                                enum ip_conntrack_info ctinfo,
 835                                struct nf_conntrack_tuple_hash *h)
 836 {
 837         /* This is the conntrack entry already in hashes that won race. */
 838         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 839         const struct nf_conntrack_l4proto *l4proto;
 840         enum ip_conntrack_info oldinfo;
 841         struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo);
 842
 843         l4proto = __nf_ct_l4proto_find(nf_ct_protonum(ct));
 844         if (l4proto->allow_clash &&
 845             !nf_ct_is_dying(ct) &&
 846             atomic_inc_not_zero(&ct->ct_general.use)) {
 847                 if (((ct->status & IPS_NAT_DONE_MASK) == 0) ||
 848                     nf_ct_match(ct, loser_ct)) {
 849                         nf_ct_acct_merge(ct, ctinfo, loser_ct);
 850                         nf_conntrack_put(&loser_ct->ct_general);
 851                         nf_ct_set(skb, ct, oldinfo);
 852                         return NF_ACCEPT;
 853                 }
 854                 nf_ct_put(ct);
 855         }
 856         NF_CT_STAT_INC(net, drop);
 857         return NF_DROP;
 858 }
 859
 860 /* Confirm a connection given skb; places it in hash table */
 861 int
 862 __nf_conntrack_confirm(struct sk_buff *skb)
 863 {
 864         const struct nf_conntrack_zone *zone;
 865         unsigned int hash, reply_hash;
 866         struct nf_conntrack_tuple_hash *h;
 867         struct nf_conn *ct;
 868         struct nf_conn_help *help;
 869         struct nf_conn_tstamp *tstamp;
 870         struct hlist_nulls_node *n;
 871         enum ip_conntrack_info ctinfo;
 872         struct net *net;
 873         unsigned int sequence;
 874         int ret = NF_DROP;
 875
 876         ct = nf_ct_get(skb, &ctinfo);
 877         net = nf_ct_net(ct);
 878
 879         /* ipt_REJECT uses nf_conntrack_attach to attach related
 880            ICMP/TCP RST packets in other direction.  Actual packet
 881            which created connection will be IP_CT_NEW or for an
 882            expected connection, IP_CT_RELATED. */
 883         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 884                 return NF_ACCEPT;
 885
 886         zone = nf_ct_zone(ct);
 887         local_bh_disable();
 888
 889         do {
 890                 sequence = read_seqcount_begin(&nf_conntrack_generation);
 891                 /* reuse the hash saved before */
 892                 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 893                 hash = scale_hash(hash);
 894                 reply_hash = hash_conntrack(net,
 895                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 896
 897         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 898
 899         /* We're not in hash table, and we refuse to set up related
 900          * connections for unconfirmed conns.  But packet copies and
 901          * REJECT will give spurious warnings here.
 902          */
 903
 904         /* No external references means no one else could have
 905          * confirmed us.
 906          */
 907         WARN_ON(nf_ct_is_confirmed(ct));
 908         pr_debug("Confirming conntrack %p\n", ct);
 909         /* We have to check the DYING flag after unlink to prevent
 910          * a race against nf_ct_get_next_corpse() possibly called from
 911          * user context, else we insert an already 'dead' hash, blocking
 912          * further use of that particular connection -JM.
 913          */
 914         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 915
 916         if (unlikely(nf_ct_is_dying(ct))) {
 917                 nf_ct_add_to_dying_list(ct);
 918                 goto dying;
 919         }
 920
 921         /* See if there's one in the list already, including reverse:
 922            NAT could have grabbed it without realizing, since we're
 923            not in the hash.  If there is, we lost race. */
 924         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
 925                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 926                                     zone, net))
 927                         goto out;
 928
 929         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
 930                 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 931                                     zone, net))
 932                         goto out;
 933
 934         /* Timer relative to confirmation time, not original
 935            setting time, otherwise we'd get timer wrap in
 936            weird delay cases. */
 937         ct->timeout += nfct_time_stamp;
 938         atomic_inc(&ct->ct_general.use);
 939         ct->status |= IPS_CONFIRMED;
 940
 941         /* set conntrack timestamp, if enabled. */
 942         tstamp = nf_conn_tstamp_find(ct);
 943         if (tstamp) {
 944                 if (skb->tstamp == 0)
 945                         __net_timestamp(skb);
 946
 947                 tstamp->start = ktime_to_ns(skb->tstamp);
 948         }
 949         /* Since the lookup is lockless, hash insertion must be done after
 950          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 951          * guarantee that no other CPU can find the conntrack before the above
 952          * stores are visible.
 953          */
 954         __nf_conntrack_hash_insert(ct, hash, reply_hash);
 955         nf_conntrack_double_unlock(hash, reply_hash);
 956         local_bh_enable();
 957
 958         help = nfct_help(ct);
 959         if (help && help->helper)
 960                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 961
 962         nf_conntrack_event_cache(master_ct(ct) ?
 963                                  IPCT_RELATED : IPCT_NEW, ct);
 964         return NF_ACCEPT;
 965
 966 out:
 967         nf_ct_add_to_dying_list(ct);
 968         ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
 969 dying:
 970         nf_conntrack_double_unlock(hash, reply_hash);
 971         NF_CT_STAT_INC(net, insert_failed);
 972         local_bh_enable();
 973         return ret;
 974 }
 975 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 976
 977 /* Returns true if a connection correspondings to the tuple (required
 978    for NAT). */
 979 int
 980 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 981                          const struct nf_conn *ignored_conntrack)
 982 {
 983         struct net *net = nf_ct_net(ignored_conntrack);
 984         const struct nf_conntrack_zone *zone;
 985         struct nf_conntrack_tuple_hash *h;
 986         struct hlist_nulls_head *ct_hash;
 987         unsigned int hash, hsize;
 988         struct hlist_nulls_node *n;
 989         struct nf_conn *ct;
 990
 991         zone = nf_ct_zone(ignored_conntrack);
 992
 993         rcu_read_lock();
 994  begin:
 995         nf_conntrack_get_ht(&ct_hash, &hsize);
 996         hash = __hash_conntrack(net, tuple, hsize);
 997
 998         hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
 999                 ct = nf_ct_tuplehash_to_ctrack(h);
1000
1001                 if (ct == ignored_conntrack)
1002                         continue;
1003
1004                 if (nf_ct_is_expired(ct)) {
1005                         nf_ct_gc_expired(ct);
1006                         continue;
1007                 }
1008
1009                 if (nf_ct_key_equal(h, tuple, zone, net)) {
1010                         /* Tuple is taken already, so caller will need to find
1011                          * a new source port to use.
1012                          *
1013                          * Only exception:
1014                          * If the *original tuples* are identical, then both
1015                          * conntracks refer to the same flow.
1016                          * This is a rare situation, it can occur e.g. when
1017                          * more than one UDP packet is sent from same socket
1018                          * in different threads.
1019                          *
1020                          * Let nf_ct_resolve_clash() deal with this later.
1021                          */
1022                         if (nf_ct_tuple_equal(&ignored_conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1023                                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple))
1024                                 continue;
1025
1026                         NF_CT_STAT_INC_ATOMIC(net, found);
1027                         rcu_read_unlock();
1028                         return 1;
1029                 }
1030         }
1031
1032         if (get_nulls_value(n) != hash) {
1033                 NF_CT_STAT_INC_ATOMIC(net, search_restart);
1034                 goto begin;
1035         }
1036
1037         rcu_read_unlock();
1038
1039         return 0;
1040 }
1041 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
1042
1043 #define NF_CT_EVICTION_RANGE    8
1044
1045 /* There's a small race here where we may free a just-assured
1046    connection.  Too bad: we're in trouble anyway. */
1047 static unsigned int early_drop_list(struct net *net,
1048                                     struct hlist_nulls_head *head)
1049 {
1050         struct nf_conntrack_tuple_hash *h;
1051         struct hlist_nulls_node *n;
1052         unsigned int drops = 0;
1053         struct nf_conn *tmp;
1054
1055         hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
1056                 tmp = nf_ct_tuplehash_to_ctrack(h);
1057
1058                 if (test_bit(IPS_OFFLOAD_BIT, &tmp->status))
1059                         continue;
1060
1061                 if (nf_ct_is_expired(tmp)) {
1062                         nf_ct_gc_expired(tmp);
1063                         continue;
1064                 }
1065
1066                 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
1067                     !net_eq(nf_ct_net(tmp), net) ||
1068                     nf_ct_is_dying(tmp))
1069                         continue;
1070
1071                 if (!atomic_inc_not_zero(&tmp->ct_general.use))
1072                         continue;
1073
1074                 /* kill only if still in same netns -- might have moved due to
1075                  * SLAB_TYPESAFE_BY_RCU rules.
1076                  *
1077                  * We steal the timer reference.  If that fails timer has
1078                  * already fired or someone else deleted it. Just drop ref
1079                  * and move to next entry.
1080                  */
1081                 if (net_eq(nf_ct_net(tmp), net) &&
1082                     nf_ct_is_confirmed(tmp) &&
1083                     nf_ct_delete(tmp, 0, 0))
1084                         drops++;
1085
1086                 nf_ct_put(tmp);
1087         }
1088
1089         return drops;
1090 }
1091
1092 static noinline int early_drop(struct net *net, unsigned int hash)
1093 {
1094         unsigned int i, bucket;
1095
1096         for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
1097                 struct hlist_nulls_head *ct_hash;
1098                 unsigned int hsize, drops;
1099
1100                 rcu_read_lock();
1101                 nf_conntrack_get_ht(&ct_hash, &hsize);
1102                 if (!i)
1103                         bucket = reciprocal_scale(hash, hsize);
1104                 else
1105                         bucket = (bucket + 1) % hsize;
1106
1107                 drops = early_drop_list(net, &ct_hash[bucket]);
1108                 rcu_read_unlock();
1109
1110                 if (drops) {
1111                         NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
1112                         return true;
1113                 }
1114         }
1115
1116         return false;
1117 }
1118
1119 static bool gc_worker_skip_ct(const struct nf_conn *ct)
1120 {
1121         return !nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct);
1122 }
1123
1124 static bool gc_worker_can_early_drop(const struct nf_conn *ct)
1125 {
1126         const struct nf_conntrack_l4proto *l4proto;
1127
1128         if (!test_bit(IPS_ASSURED_BIT, &ct->status))
1129                 return true;
1130
1131         l4proto = __nf_ct_l4proto_find(nf_ct_protonum(ct));
1132         if (l4proto->can_early_drop && l4proto->can_early_drop(ct))
1133                 return true;
1134
1135         return false;
1136 }
1137
1138 #define DAY     (86400 * HZ)
1139
1140 /* Set an arbitrary timeout large enough not to ever expire, this save
1141  * us a check for the IPS_OFFLOAD_BIT from the packet path via
1142  * nf_ct_is_expired().
1143  */
1144 static void nf_ct_offload_timeout(struct nf_conn *ct)
1145 {
1146         if (nf_ct_expires(ct) < DAY / 2)
1147                 ct->timeout = nfct_time_stamp + DAY;
1148 }
1149
1150 static void gc_worker(struct work_struct *work)
1151 {
1152         unsigned int min_interval = max(HZ / GC_MAX_BUCKETS_DIV, 1u);
1153         unsigned int i, goal, buckets = 0, expired_count = 0;
1154         unsigned int nf_conntrack_max95 = 0;
1155         struct conntrack_gc_work *gc_work;
1156         unsigned int ratio, scanned = 0;
1157         unsigned long next_run;
1158
1159         gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
1160
1161         goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV;
1162         i = gc_work->last_bucket;
1163         if (gc_work->early_drop)
1164                 nf_conntrack_max95 = nf_conntrack_max / 100u * 95u;
1165
1166         do {
1167                 struct nf_conntrack_tuple_hash *h;
1168                 struct hlist_nulls_head *ct_hash;
1169                 struct hlist_nulls_node *n;
1170                 unsigned int hashsz;
1171                 struct nf_conn *tmp;
1172
1173                 i++;
1174                 rcu_read_lock();
1175
1176                 nf_conntrack_get_ht(&ct_hash, &hashsz);
1177                 if (i >= hashsz)
1178                         i = 0;
1179
1180                 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
1181                         struct net *net;
1182
1183                         tmp = nf_ct_tuplehash_to_ctrack(h);
1184
1185                         scanned++;
1186                         if (test_bit(IPS_OFFLOAD_BIT, &tmp->status)) {
1187                                 nf_ct_offload_timeout(tmp);
1188                                 continue;
1189                         }
1190
1191                         if (nf_ct_is_expired(tmp)) {
1192                                 nf_ct_gc_expired(tmp);
1193                                 expired_count++;
1194                                 continue;
1195                         }
1196
1197                         if (nf_conntrack_max95 == 0 || gc_worker_skip_ct(tmp))
1198                                 continue;
1199
1200                         net = nf_ct_net(tmp);
1201                         if (atomic_read(&net->ct.count) < nf_conntrack_max95)
1202                                 continue;
1203
1204                         /* need to take reference to avoid possible races */
1205                         if (!atomic_inc_not_zero(&tmp->ct_general.use))
1206                                 continue;
1207
1208                         if (gc_worker_skip_ct(tmp)) {
1209                                 nf_ct_put(tmp);
1210                                 continue;
1211                         }
1212
1213                         if (gc_worker_can_early_drop(tmp))
1214                                 nf_ct_kill(tmp);
1215
1216                         nf_ct_put(tmp);
1217                 }
1218
1219                 /* could check get_nulls_value() here and restart if ct
1220                  * was moved to another chain.  But given gc is best-effort
1221                  * we will just continue with next hash slot.
1222                  */
1223                 rcu_read_unlock();
1224                 cond_resched();
1225         } while (++buckets < goal);
1226
1227         if (gc_work->exiting)
1228                 return;
1229
1230         /*
1231          * Eviction will normally happen from the packet path, and not
1232          * from this gc worker.
1233          *
1234          * This worker is only here to reap expired entries when system went
1235          * idle after a busy period.
1236          *
1237          * The heuristics below are supposed to balance conflicting goals:
1238          *
1239          * 1. Minimize time until we notice a stale entry
1240          * 2. Maximize scan intervals to not waste cycles
1241          *
1242          * Normally, expire ratio will be close to 0.
1243          *
1244          * As soon as a sizeable fraction of the entries have expired
1245          * increase scan frequency.
1246          */
1247         ratio = scanned ? expired_count * 100 / scanned : 0;
1248         if (ratio > GC_EVICT_RATIO) {
1249                 gc_work->next_gc_run = min_interval;
1250         } else {
1251                 unsigned int max = GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV;
1252
1253                 BUILD_BUG_ON((GC_MAX_SCAN_JIFFIES / GC_MAX_BUCKETS_DIV) == 0);
1254
1255                 gc_work->next_gc_run += min_interval;
1256                 if (gc_work->next_gc_run > max)
1257                         gc_work->next_gc_run = max;
1258         }
1259
1260         next_run = gc_work->next_gc_run;
1261         gc_work->last_bucket = i;
1262         gc_work->early_drop = false;
1263         queue_delayed_work(system_power_efficient_wq, &gc_work->dwork, next_run);
1264 }
1265
1266 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1267 {
1268         INIT_DEFERRABLE_WORK(&gc_work->dwork, gc_worker);
1269         gc_work->next_gc_run = HZ;
1270         gc_work->exiting = false;
1271 }
1272
1273 static struct nf_conn *
1274 __nf_conntrack_alloc(struct net *net,
1275                      const struct nf_conntrack_zone *zone,
1276                      const struct nf_conntrack_tuple *orig,
1277                      const struct nf_conntrack_tuple *repl,
1278                      gfp_t gfp, u32 hash)
1279 {
1280         struct nf_conn *ct;
1281
1282         /* We don't want any race condition at early drop stage */
1283         atomic_inc(&net->ct.count);
1284
1285         if (nf_conntrack_max &&
1286             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
1287                 if (!early_drop(net, hash)) {
1288                         if (!conntrack_gc_work.early_drop)
1289                                 conntrack_gc_work.early_drop = true;
1290                         atomic_dec(&net->ct.count);
1291                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1292                         return ERR_PTR(-ENOMEM);
1293                 }
1294         }
1295
1296         /*
1297          * Do not use kmem_cache_zalloc(), as this cache uses
1298          * SLAB_TYPESAFE_BY_RCU.
1299          */
1300         ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1301         if (ct == NULL)
1302                 goto out;
1303
1304         spin_lock_init(&ct->lock);
1305         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1306         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1307         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1308         /* save hash for reusing when confirming */
1309         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1310         ct->status = 0;
1311         write_pnet(&ct->ct_net, net);
1312         memset(&ct->__nfct_init_offset[0], 0,
1313                offsetof(struct nf_conn, proto) -
1314                offsetof(struct nf_conn, __nfct_init_offset[0]));
1315
1316         nf_ct_zone_add(ct, zone);
1317
1318         /* Because we use RCU lookups, we set ct_general.use to zero before
1319          * this is inserted in any list.
1320          */
1321         atomic_set(&ct->ct_general.use, 0);
1322         return ct;
1323 out:
1324         atomic_dec(&net->ct.count);
1325         return ERR_PTR(-ENOMEM);
1326 }
1327
1328 struct nf_conn *nf_conntrack_alloc(struct net *net,
1329                                    const struct nf_conntrack_zone *zone,
1330                                    const struct nf_conntrack_tuple *orig,
1331                                    const struct nf_conntrack_tuple *repl,
1332                                    gfp_t gfp)
1333 {
1334         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1335 }
1336 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1337
1338 void nf_conntrack_free(struct nf_conn *ct)
1339 {
1340         struct net *net = nf_ct_net(ct);
1341
1342         /* A freed object has refcnt == 0, that's
1343          * the golden rule for SLAB_TYPESAFE_BY_RCU
1344          */
1345         WARN_ON(atomic_read(&ct->ct_general.use) != 0);
1346
1347         nf_ct_ext_destroy(ct);
1348         nf_ct_ext_free(ct);
1349         kmem_cache_free(nf_conntrack_cachep, ct);
1350         smp_mb__before_atomic();
1351         atomic_dec(&net->ct.count);
1352 }
1353 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1354
1355
1356 /* Allocate a new conntrack: we return -ENOMEM if classification
1357    failed due to stress.  Otherwise it really is unclassifiable. */
1358 static noinline struct nf_conntrack_tuple_hash *
1359 init_conntrack(struct net *net, struct nf_conn *tmpl,
1360                const struct nf_conntrack_tuple *tuple,
1361                const struct nf_conntrack_l4proto *l4proto,
1362                struct sk_buff *skb,
1363                unsigned int dataoff, u32 hash)
1364 {
1365         struct nf_conn *ct;
1366         struct nf_conn_help *help;
1367         struct nf_conntrack_tuple repl_tuple;
1368         struct nf_conntrack_ecache *ecache;
1369         struct nf_conntrack_expect *exp = NULL;
1370         const struct nf_conntrack_zone *zone;
1371         struct nf_conn_timeout *timeout_ext;
1372         struct nf_conntrack_zone tmp;
1373
1374         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l4proto)) {
1375                 pr_debug("Can't invert tuple.\n");
1376                 return NULL;
1377         }
1378
1379         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1380         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1381                                   hash);
1382         if (IS_ERR(ct))
1383                 return (struct nf_conntrack_tuple_hash *)ct;
1384
1385         if (!nf_ct_add_synproxy(ct, tmpl)) {
1386                 nf_conntrack_free(ct);
1387                 return ERR_PTR(-ENOMEM);
1388         }
1389
1390         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1391
1392         if (timeout_ext)
1393                 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1394                                       GFP_ATOMIC);
1395
1396         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1397         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1398         nf_ct_labels_ext_add(ct);
1399
1400         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1401         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1402                                  ecache ? ecache->expmask : 0,
1403                              GFP_ATOMIC);
1404
1405         local_bh_disable();
1406         if (net->ct.expect_count) {
1407                 spin_lock(&nf_conntrack_expect_lock);
1408                 exp = nf_ct_find_expectation(net, zone, tuple);
1409                 if (exp) {
1410                         pr_debug("expectation arrives ct=%p exp=%p\n",
1411                                  ct, exp);
1412                         /* Welcome, Mr. Bond.  We've been expecting you... */
1413                         __set_bit(IPS_EXPECTED_BIT, &ct->status);
1414                         /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1415                         ct->master = exp->master;
1416                         if (exp->helper) {
1417                                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
1418                                 if (help)
1419                                         rcu_assign_pointer(help->helper, exp->helper);
1420                         }
1421
1422 #ifdef CONFIG_NF_CONNTRACK_MARK
1423                         ct->mark = exp->master->mark;
1424 #endif
1425 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1426                         ct->secmark = exp->master->secmark;
1427 #endif
1428                         NF_CT_STAT_INC(net, expect_new);
1429                 }
1430                 spin_unlock(&nf_conntrack_expect_lock);
1431         }
1432         if (!exp)
1433                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1434
1435         /* Now it is inserted into the unconfirmed list, bump refcount */
1436         nf_conntrack_get(&ct->ct_general);
1437         nf_ct_add_to_unconfirmed_list(ct);
1438
1439         local_bh_enable();
1440
1441         if (exp) {
1442                 if (exp->expectfn)
1443                         exp->expectfn(ct, exp);
1444                 nf_ct_expect_put(exp);
1445         }
1446
1447         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1448 }
1449
1450 /* On success, returns 0, sets skb->_nfct | ctinfo */
1451 static int
1452 resolve_normal_ct(struct nf_conn *tmpl,
1453                   struct sk_buff *skb,
1454                   unsigned int dataoff,
1455                   u_int8_t protonum,
1456                   const struct nf_conntrack_l4proto *l4proto,
1457                   const struct nf_hook_state *state)
1458 {
1459         const struct nf_conntrack_zone *zone;
1460         struct nf_conntrack_tuple tuple;
1461         struct nf_conntrack_tuple_hash *h;
1462         enum ip_conntrack_info ctinfo;
1463         struct nf_conntrack_zone tmp;
1464         struct nf_conn *ct;
1465         u32 hash;
1466
1467         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1468                              dataoff, state->pf, protonum, state->net,
1469                              &tuple, l4proto)) {
1470                 pr_debug("Can't get tuple\n");
1471                 return 0;
1472         }
1473
1474         /* look for tuple match */
1475         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1476         hash = hash_conntrack_raw(&tuple, state->net);
1477         h = __nf_conntrack_find_get(state->net, zone, &tuple, hash);
1478         if (!h) {
1479                 h = init_conntrack(state->net, tmpl, &tuple, l4proto,
1480                                    skb, dataoff, hash);
1481                 if (!h)
1482                         return 0;
1483                 if (IS_ERR(h))
1484                         return PTR_ERR(h);
1485         }
1486         ct = nf_ct_tuplehash_to_ctrack(h);
1487
1488         /* It exists; we have (non-exclusive) reference. */
1489         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1490                 ctinfo = IP_CT_ESTABLISHED_REPLY;
1491         } else {
1492                 /* Once we've had two way comms, always ESTABLISHED. */
1493                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1494                         pr_debug("normal packet for %p\n", ct);
1495                         ctinfo = IP_CT_ESTABLISHED;
1496                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1497                         pr_debug("related packet for %p\n", ct);
1498                         ctinfo = IP_CT_RELATED;
1499                 } else {
1500                         pr_debug("new packet for %p\n", ct);
1501                         ctinfo = IP_CT_NEW;
1502                 }
1503         }
1504         nf_ct_set(skb, ct, ctinfo);
1505         return 0;
1506 }
1507
1508 /*
1509  * icmp packets need special treatment to handle error messages that are
1510  * related to a connection.
1511  *
1512  * Callers need to check if skb has a conntrack assigned when this
1513  * helper returns; in such case skb belongs to an already known connection.
1514  */
1515 static unsigned int __cold
1516 nf_conntrack_handle_icmp(struct nf_conn *tmpl,
1517                          struct sk_buff *skb,
1518                          unsigned int dataoff,
1519                          u8 protonum,
1520                          const struct nf_hook_state *state)
1521 {
1522         int ret;
1523
1524         if (state->pf == NFPROTO_IPV4 && protonum == IPPROTO_ICMP)
1525                 ret = nf_conntrack_icmpv4_error(tmpl, skb, dataoff, state);
1526 #if IS_ENABLED(CONFIG_IPV6)
1527         else if (state->pf == NFPROTO_IPV6 && protonum == IPPROTO_ICMPV6)
1528                 ret = nf_conntrack_icmpv6_error(tmpl, skb, dataoff, state);
1529 #endif
1530         else
1531                 return NF_ACCEPT;
1532
1533         if (ret <= 0) {
1534                 NF_CT_STAT_INC_ATOMIC(state->net, error);
1535                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1536         }
1537
1538         return ret;
1539 }
1540
1541 unsigned int
1542 nf_conntrack_in(struct sk_buff *skb, const struct nf_hook_state *state)
1543 {
1544         const struct nf_conntrack_l4proto *l4proto;
1545         enum ip_conntrack_info ctinfo;
1546         struct nf_conn *ct, *tmpl;
1547         u_int8_t protonum;
1548         int dataoff, ret;
1549
1550         tmpl = nf_ct_get(skb, &ctinfo);
1551         if (tmpl || ctinfo == IP_CT_UNTRACKED) {
1552                 /* Previously seen (loopback or untracked)?  Ignore. */
1553                 if ((tmpl && !nf_ct_is_template(tmpl)) ||
1554                      ctinfo == IP_CT_UNTRACKED) {
1555                         NF_CT_STAT_INC_ATOMIC(state->net, ignore);
1556                         return NF_ACCEPT;
1557                 }
1558                 skb->_nfct = 0;
1559         }
1560
1561         /* rcu_read_lock()ed by nf_hook_thresh */
1562         dataoff = get_l4proto(skb, skb_network_offset(skb), state->pf, &protonum);
1563         if (dataoff <= 0) {
1564                 pr_debug("not prepared to track yet or error occurred\n");
1565                 NF_CT_STAT_INC_ATOMIC(state->net, error);
1566                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1567                 ret = NF_ACCEPT;
1568                 goto out;
1569         }
1570
1571         l4proto = __nf_ct_l4proto_find(protonum);
1572
1573         if (protonum == IPPROTO_ICMP || protonum == IPPROTO_ICMPV6) {
1574                 ret = nf_conntrack_handle_icmp(tmpl, skb, dataoff,
1575                                                protonum, state);
1576                 if (ret <= 0) {
1577                         ret = -ret;
1578                         goto out;
1579                 }
1580                 /* ICMP[v6] protocol trackers may assign one conntrack. */
1581                 if (skb->_nfct)
1582                         goto out;
1583         }
1584 repeat:
1585         ret = resolve_normal_ct(tmpl, skb, dataoff,
1586                                 protonum, l4proto, state);
1587         if (ret < 0) {
1588                 /* Too stressed to deal. */
1589                 NF_CT_STAT_INC_ATOMIC(state->net, drop);
1590                 ret = NF_DROP;
1591                 goto out;
1592         }
1593
1594         ct = nf_ct_get(skb, &ctinfo);
1595         if (!ct) {
1596                 /* Not valid part of a connection */
1597                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1598                 ret = NF_ACCEPT;
1599                 goto out;
1600         }
1601
1602         ret = l4proto->packet(ct, skb, dataoff, ctinfo, state);
1603         if (ret <= 0) {
1604                 /* Invalid: inverse of the return code tells
1605                  * the netfilter core what to do */
1606                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1607                 nf_conntrack_put(&ct->ct_general);
1608                 skb->_nfct = 0;
1609                 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1610                 if (ret == -NF_DROP)
1611                         NF_CT_STAT_INC_ATOMIC(state->net, drop);
1612                 /* Special case: TCP tracker reports an attempt to reopen a
1613                  * closed/aborted connection. We have to go back and create a
1614                  * fresh conntrack.
1615                  */
1616                 if (ret == -NF_REPEAT)
1617                         goto repeat;
1618                 ret = -ret;
1619                 goto out;
1620         }
1621
1622         if (ctinfo == IP_CT_ESTABLISHED_REPLY &&
1623             !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1624                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1625 out:
1626         if (tmpl)
1627                 nf_ct_put(tmpl);
1628
1629         return ret;
1630 }
1631 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1632
1633 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1634                           const struct nf_conntrack_tuple *orig)
1635 {
1636         bool ret;
1637
1638         rcu_read_lock();
1639         ret = nf_ct_invert_tuple(inverse, orig,
1640                                  __nf_ct_l4proto_find(orig->dst.protonum));
1641         rcu_read_unlock();
1642         return ret;
1643 }
1644 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1645
1646 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1647    implicitly racy: see __nf_conntrack_confirm */
1648 void nf_conntrack_alter_reply(struct nf_conn *ct,
1649                               const struct nf_conntrack_tuple *newreply)
1650 {
1651         struct nf_conn_help *help = nfct_help(ct);
1652
1653         /* Should be unconfirmed, so not in hash table yet */
1654         WARN_ON(nf_ct_is_confirmed(ct));
1655
1656         pr_debug("Altering reply tuple of %p to ", ct);
1657         nf_ct_dump_tuple(newreply);
1658
1659         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1660         if (ct->master || (help && !hlist_empty(&help->expectations)))
1661                 return;
1662
1663         rcu_read_lock();
1664         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1665         rcu_read_unlock();
1666 }
1667 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1668
1669 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1670 void __nf_ct_refresh_acct(struct nf_conn *ct,
1671                           enum ip_conntrack_info ctinfo,
1672                           const struct sk_buff *skb,
1673                           unsigned long extra_jiffies,
1674                           int do_acct)
1675 {
1676         WARN_ON(!skb);
1677
1678         /* Only update if this is not a fixed timeout */
1679         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1680                 goto acct;
1681
1682         /* If not in hash table, timer will not be active yet */
1683         if (nf_ct_is_confirmed(ct))
1684                 extra_jiffies += nfct_time_stamp;
1685
1686         ct->timeout = extra_jiffies;
1687 acct:
1688         if (do_acct)
1689                 nf_ct_acct_update(ct, ctinfo, skb->len);
1690 }
1691 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1692
1693 bool nf_ct_kill_acct(struct nf_conn *ct,
1694                      enum ip_conntrack_info ctinfo,
1695                      const struct sk_buff *skb)
1696 {
1697         nf_ct_acct_update(ct, ctinfo, skb->len);
1698
1699         return nf_ct_delete(ct, 0, 0);
1700 }
1701 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1702
1703 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1704
1705 #include <linux/netfilter/nfnetlink.h>
1706 #include <linux/netfilter/nfnetlink_conntrack.h>
1707 #include <linux/mutex.h>
1708
1709 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1710  * in ip_conntrack_core, since we don't want the protocols to autoload
1711  * or depend on ctnetlink */
1712 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1713                                const struct nf_conntrack_tuple *tuple)
1714 {
1715         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1716             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1717                 goto nla_put_failure;
1718         return 0;
1719
1720 nla_put_failure:
1721         return -1;
1722 }
1723 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1724
1725 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1726         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1727         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1728 };
1729 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1730
1731 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1732                                struct nf_conntrack_tuple *t)
1733 {
1734         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1735                 return -EINVAL;
1736
1737         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1738         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1739
1740         return 0;
1741 }
1742 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1743
1744 unsigned int nf_ct_port_nlattr_tuple_size(void)
1745 {
1746         static unsigned int size __read_mostly;
1747
1748         if (!size)
1749                 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1750
1751         return size;
1752 }
1753 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1754 #endif
1755
1756 /* Used by ipt_REJECT and ip6t_REJECT. */
1757 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1758 {
1759         struct nf_conn *ct;
1760         enum ip_conntrack_info ctinfo;
1761
1762         /* This ICMP is in reverse direction to the packet which caused it */
1763         ct = nf_ct_get(skb, &ctinfo);
1764         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1765                 ctinfo = IP_CT_RELATED_REPLY;
1766         else
1767                 ctinfo = IP_CT_RELATED;
1768
1769         /* Attach to new skbuff, and increment count */
1770         nf_ct_set(nskb, ct, ctinfo);
1771         nf_conntrack_get(skb_nfct(nskb));
1772 }
1773
1774 static int nf_conntrack_update(struct net *net, struct sk_buff *skb)
1775 {
1776         const struct nf_conntrack_l4proto *l4proto;
1777         struct nf_conntrack_tuple_hash *h;
1778         struct nf_conntrack_tuple tuple;
1779         enum ip_conntrack_info ctinfo;
1780         struct nf_nat_hook *nat_hook;
1781         unsigned int status;
1782         struct nf_conn *ct;
1783         int dataoff;
1784         u16 l3num;
1785         u8 l4num;
1786
1787         ct = nf_ct_get(skb, &ctinfo);
1788         if (!ct || nf_ct_is_confirmed(ct))
1789                 return 0;
1790
1791         l3num = nf_ct_l3num(ct);
1792
1793         dataoff = get_l4proto(skb, skb_network_offset(skb), l3num, &l4num);
1794         if (dataoff <= 0)
1795                 return -1;
1796
1797         l4proto = nf_ct_l4proto_find_get(l4num);
1798
1799         if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
1800                              l4num, net, &tuple, l4proto))
1801                 return -1;
1802
1803         if (ct->status & IPS_SRC_NAT) {
1804                 memcpy(tuple.src.u3.all,
1805                        ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.all,
1806                        sizeof(tuple.src.u3.all));
1807                 tuple.src.u.all =
1808                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.all;
1809         }
1810
1811         if (ct->status & IPS_DST_NAT) {
1812                 memcpy(tuple.dst.u3.all,
1813                        ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.all,
1814                        sizeof(tuple.dst.u3.all));
1815                 tuple.dst.u.all =
1816                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u.all;
1817         }
1818
1819         h = nf_conntrack_find_get(net, nf_ct_zone(ct), &tuple);
1820         if (!h)
1821                 return 0;
1822
1823         /* Store status bits of the conntrack that is clashing to re-do NAT
1824          * mangling according to what it has been done already to this packet.
1825          */
1826         status = ct->status;
1827
1828         nf_ct_put(ct);
1829         ct = nf_ct_tuplehash_to_ctrack(h);
1830         nf_ct_set(skb, ct, ctinfo);
1831
1832         nat_hook = rcu_dereference(nf_nat_hook);
1833         if (!nat_hook)
1834                 return 0;
1835
1836         if (status & IPS_SRC_NAT &&
1837             nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_SRC,
1838                                 IP_CT_DIR_ORIGINAL) == NF_DROP)
1839                 return -1;
1840
1841         if (status & IPS_DST_NAT &&
1842             nat_hook->manip_pkt(skb, ct, NF_NAT_MANIP_DST,
1843                                 IP_CT_DIR_ORIGINAL) == NF_DROP)
1844                 return -1;
1845
1846         return 0;
1847 }
1848
1849 static bool nf_conntrack_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
1850                                        const struct sk_buff *skb)
1851 {
1852         const struct nf_conntrack_tuple *src_tuple;
1853         const struct nf_conntrack_tuple_hash *hash;
1854         struct nf_conntrack_tuple srctuple;
1855         enum ip_conntrack_info ctinfo;
1856         struct nf_conn *ct;
1857
1858         ct = nf_ct_get(skb, &ctinfo);
1859         if (ct) {
1860                 src_tuple = nf_ct_tuple(ct, CTINFO2DIR(ctinfo));
1861                 memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
1862                 return true;
1863         }
1864
1865         if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
1866                                NFPROTO_IPV4, dev_net(skb->dev),
1867                                &srctuple))
1868                 return false;
1869
1870         hash = nf_conntrack_find_get(dev_net(skb->dev),
1871                                      &nf_ct_zone_dflt,
1872                                      &srctuple);
1873         if (!hash)
1874                 return false;
1875
1876         ct = nf_ct_tuplehash_to_ctrack(hash);
1877         src_tuple = nf_ct_tuple(ct, !hash->tuple.dst.dir);
1878         memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
1879         nf_ct_put(ct);
1880
1881         return true;
1882 }
1883
1884 /* Bring out ya dead! */
1885 static struct nf_conn *
1886 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
1887                 void *data, unsigned int *bucket)
1888 {
1889         struct nf_conntrack_tuple_hash *h;
1890         struct nf_conn *ct;
1891         struct hlist_nulls_node *n;
1892         spinlock_t *lockp;
1893
1894         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1895                 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1896                 local_bh_disable();
1897                 nf_conntrack_lock(lockp);
1898                 if (*bucket < nf_conntrack_htable_size) {
1899                         hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1900                                 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1901                                         continue;
1902                                 ct = nf_ct_tuplehash_to_ctrack(h);
1903                                 if (iter(ct, data))
1904                                         goto found;
1905                         }
1906                 }
1907                 spin_unlock(lockp);
1908                 local_bh_enable();
1909                 cond_resched();
1910         }
1911
1912         return NULL;
1913 found:
1914         atomic_inc(&ct->ct_general.use);
1915         spin_unlock(lockp);
1916         local_bh_enable();
1917         return ct;
1918 }
1919
1920 static void nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data),
1921                                   void *data, u32 portid, int report)
1922 {
1923         unsigned int bucket = 0, sequence;
1924         struct nf_conn *ct;
1925
1926         might_sleep();
1927
1928         for (;;) {
1929                 sequence = read_seqcount_begin(&nf_conntrack_generation);
1930
1931                 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
1932                         /* Time to push up daises... */
1933
1934                         nf_ct_delete(ct, portid, report);
1935                         nf_ct_put(ct);
1936                         cond_resched();
1937                 }
1938
1939                 if (!read_seqcount_retry(&nf_conntrack_generation, sequence))
1940                         break;
1941                 bucket = 0;
1942         }
1943 }
1944
1945 struct iter_data {
1946         int (*iter)(struct nf_conn *i, void *data);
1947         void *data;
1948         struct net *net;
1949 };
1950
1951 static int iter_net_only(struct nf_conn *i, void *data)
1952 {
1953         struct iter_data *d = data;
1954
1955         if (!net_eq(d->net, nf_ct_net(i)))
1956                 return 0;
1957
1958         return d->iter(i, d->data);
1959 }
1960
1961 static void
1962 __nf_ct_unconfirmed_destroy(struct net *net)
1963 {
1964         int cpu;
1965
1966         for_each_possible_cpu(cpu) {
1967                 struct nf_conntrack_tuple_hash *h;
1968                 struct hlist_nulls_node *n;
1969                 struct ct_pcpu *pcpu;
1970
1971                 pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1972
1973                 spin_lock_bh(&pcpu->lock);
1974                 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1975                         struct nf_conn *ct;
1976
1977                         ct = nf_ct_tuplehash_to_ctrack(h);
1978
1979                         /* we cannot call iter() on unconfirmed list, the
1980                          * owning cpu can reallocate ct->ext at any time.
1981                          */
1982                         set_bit(IPS_DYING_BIT, &ct->status);
1983                 }
1984                 spin_unlock_bh(&pcpu->lock);
1985                 cond_resched();
1986         }
1987 }
1988
1989 void nf_ct_unconfirmed_destroy(struct net *net)
1990 {
1991         might_sleep();
1992
1993         if (atomic_read(&net->ct.count) > 0) {
1994                 __nf_ct_unconfirmed_destroy(net);
1995                 nf_queue_nf_hook_drop(net);
1996                 synchronize_net();
1997         }
1998 }
1999 EXPORT_SYMBOL_GPL(nf_ct_unconfirmed_destroy);
2000
2001 void nf_ct_iterate_cleanup_net(struct net *net,
2002                                int (*iter)(struct nf_conn *i, void *data),
2003                                void *data, u32 portid, int report)
2004 {
2005         struct iter_data d;
2006
2007         might_sleep();
2008
2009         if (atomic_read(&net->ct.count) == 0)
2010                 return;
2011
2012         d.iter = iter;
2013         d.data = data;
2014         d.net = net;
2015
2016         nf_ct_iterate_cleanup(iter_net_only, &d, portid, report);
2017 }
2018 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net);
2019
2020 /**
2021  * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
2022  * @iter: callback to invoke for each conntrack
2023  * @data: data to pass to @iter
2024  *
2025  * Like nf_ct_iterate_cleanup, but first marks conntracks on the
2026  * unconfirmed list as dying (so they will not be inserted into
2027  * main table).
2028  *
2029  * Can only be called in module exit path.
2030  */
2031 void
2032 nf_ct_iterate_destroy(int (*iter)(struct nf_conn *i, void *data), void *data)
2033 {
2034         struct net *net;
2035
2036         down_read(&net_rwsem);
2037         for_each_net(net) {
2038                 if (atomic_read(&net->ct.count) == 0)
2039                         continue;
2040                 __nf_ct_unconfirmed_destroy(net);
2041                 nf_queue_nf_hook_drop(net);
2042         }
2043         up_read(&net_rwsem);
2044
2045         /* Need to wait for netns cleanup worker to finish, if its
2046          * running -- it might have deleted a net namespace from
2047          * the global list, so our __nf_ct_unconfirmed_destroy() might
2048          * not have affected all namespaces.
2049          */
2050         net_ns_barrier();
2051
2052         /* a conntrack could have been unlinked from unconfirmed list
2053          * before we grabbed pcpu lock in __nf_ct_unconfirmed_destroy().
2054          * This makes sure its inserted into conntrack table.
2055          */
2056         synchronize_net();
2057
2058         nf_ct_iterate_cleanup(iter, data, 0, 0);
2059 }
2060 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy);
2061
2062 static int kill_all(struct nf_conn *i, void *data)
2063 {
2064         return net_eq(nf_ct_net(i), data);
2065 }
2066
2067 void nf_conntrack_cleanup_start(void)
2068 {
2069         conntrack_gc_work.exiting = true;
2070         RCU_INIT_POINTER(ip_ct_attach, NULL);
2071 }
2072
2073 void nf_conntrack_cleanup_end(void)
2074 {
2075         RCU_INIT_POINTER(nf_ct_hook, NULL);
2076         cancel_delayed_work_sync(&conntrack_gc_work.dwork);
2077         kvfree(nf_conntrack_hash);
2078
2079         nf_conntrack_proto_fini();
2080         nf_conntrack_seqadj_fini();
2081         nf_conntrack_labels_fini();
2082         nf_conntrack_helper_fini();
2083         nf_conntrack_timeout_fini();
2084         nf_conntrack_ecache_fini();
2085         nf_conntrack_tstamp_fini();
2086         nf_conntrack_acct_fini();
2087         nf_conntrack_expect_fini();
2088
2089         kmem_cache_destroy(nf_conntrack_cachep);
2090 }
2091
2092 /*
2093  * Mishearing the voices in his head, our hero wonders how he's
2094  * supposed to kill the mall.
2095  */
2096 void nf_conntrack_cleanup_net(struct net *net)
2097 {
2098         LIST_HEAD(single);
2099
2100         list_add(&net->exit_list, &single);
2101         nf_conntrack_cleanup_net_list(&single);
2102 }
2103
2104 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
2105 {
2106         int busy;
2107         struct net *net;
2108
2109         /*
2110          * This makes sure all current packets have passed through
2111          *  netfilter framework.  Roll on, two-stage module
2112          *  delete...
2113          */
2114         synchronize_net();
2115 i_see_dead_people:
2116         busy = 0;
2117         list_for_each_entry(net, net_exit_list, exit_list) {
2118                 nf_ct_iterate_cleanup(kill_all, net, 0, 0);
2119                 if (atomic_read(&net->ct.count) != 0)
2120                         busy = 1;
2121         }
2122         if (busy) {
2123                 schedule();
2124                 goto i_see_dead_people;
2125         }
2126
2127         list_for_each_entry(net, net_exit_list, exit_list) {
2128                 nf_conntrack_proto_pernet_fini(net);
2129                 nf_conntrack_ecache_pernet_fini(net);
2130                 nf_conntrack_expect_pernet_fini(net);
2131                 free_percpu(net->ct.stat);
2132                 free_percpu(net->ct.pcpu_lists);
2133         }
2134 }
2135
2136 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
2137 {
2138         struct hlist_nulls_head *hash;
2139         unsigned int nr_slots, i;
2140
2141         if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
2142                 return NULL;
2143
2144         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
2145         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
2146
2147         hash = kvmalloc_array(nr_slots, sizeof(struct hlist_nulls_head),
2148                               GFP_KERNEL | __GFP_ZERO);
2149
2150         if (hash && nulls)
2151                 for (i = 0; i < nr_slots; i++)
2152                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
2153
2154         return hash;
2155 }
2156 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
2157
2158 int nf_conntrack_hash_resize(unsigned int hashsize)
2159 {
2160         int i, bucket;
2161         unsigned int old_size;
2162         struct hlist_nulls_head *hash, *old_hash;
2163         struct nf_conntrack_tuple_hash *h;
2164         struct nf_conn *ct;
2165
2166         if (!hashsize)
2167                 return -EINVAL;
2168
2169         hash = nf_ct_alloc_hashtable(&hashsize, 1);
2170         if (!hash)
2171                 return -ENOMEM;
2172
2173         old_size = nf_conntrack_htable_size;
2174         if (old_size == hashsize) {
2175                 kvfree(hash);
2176                 return 0;
2177         }
2178
2179         local_bh_disable();
2180         nf_conntrack_all_lock();
2181         write_seqcount_begin(&nf_conntrack_generation);
2182
2183         /* Lookups in the old hash might happen in parallel, which means we
2184          * might get false negatives during connection lookup. New connections
2185          * created because of a false negative won't make it into the hash
2186          * though since that required taking the locks.
2187          */
2188
2189         for (i = 0; i < nf_conntrack_htable_size; i++) {
2190                 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
2191                         h = hlist_nulls_entry(nf_conntrack_hash[i].first,
2192                                               struct nf_conntrack_tuple_hash, hnnode);
2193                         ct = nf_ct_tuplehash_to_ctrack(h);
2194                         hlist_nulls_del_rcu(&h->hnnode);
2195                         bucket = __hash_conntrack(nf_ct_net(ct),
2196                                                   &h->tuple, hashsize);
2197                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
2198                 }
2199         }
2200         old_size = nf_conntrack_htable_size;
2201         old_hash = nf_conntrack_hash;
2202
2203         nf_conntrack_hash = hash;
2204         nf_conntrack_htable_size = hashsize;
2205
2206         write_seqcount_end(&nf_conntrack_generation);
2207         nf_conntrack_all_unlock();
2208         local_bh_enable();
2209
2210         synchronize_net();
2211         kvfree(old_hash);
2212         return 0;
2213 }
2214
2215 int nf_conntrack_set_hashsize(const char *val, const struct kernel_param *kp)
2216 {
2217         unsigned int hashsize;
2218         int rc;
2219
2220         if (current->nsproxy->net_ns != &init_net)
2221                 return -EOPNOTSUPP;
2222
2223         /* On boot, we can set this without any fancy locking. */
2224         if (!nf_conntrack_hash)
2225                 return param_set_uint(val, kp);
2226
2227         rc = kstrtouint(val, 0, &hashsize);
2228         if (rc)
2229                 return rc;
2230
2231         return nf_conntrack_hash_resize(hashsize);
2232 }
2233 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
2234
2235 static __always_inline unsigned int total_extension_size(void)
2236 {
2237         /* remember to add new extensions below */
2238         BUILD_BUG_ON(NF_CT_EXT_NUM > 9);
2239
2240         return sizeof(struct nf_ct_ext) +
2241                sizeof(struct nf_conn_help)
2242 #if IS_ENABLED(CONFIG_NF_NAT)
2243                 + sizeof(struct nf_conn_nat)
2244 #endif
2245                 + sizeof(struct nf_conn_seqadj)
2246                 + sizeof(struct nf_conn_acct)
2247 #ifdef CONFIG_NF_CONNTRACK_EVENTS
2248                 + sizeof(struct nf_conntrack_ecache)
2249 #endif
2250 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
2251                 + sizeof(struct nf_conn_tstamp)
2252 #endif
2253 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
2254                 + sizeof(struct nf_conn_timeout)
2255 #endif
2256 #ifdef CONFIG_NF_CONNTRACK_LABELS
2257                 + sizeof(struct nf_conn_labels)
2258 #endif
2259 #if IS_ENABLED(CONFIG_NETFILTER_SYNPROXY)
2260                 + sizeof(struct nf_conn_synproxy)
2261 #endif
2262         ;
2263 };
2264
2265 int nf_conntrack_init_start(void)
2266 {
2267         unsigned long nr_pages = totalram_pages();
2268         int max_factor = 8;
2269         int ret = -ENOMEM;
2270         int i;
2271
2272         /* struct nf_ct_ext uses u8 to store offsets/size */
2273         BUILD_BUG_ON(total_extension_size() > 255u);
2274
2275         seqcount_init(&nf_conntrack_generation);
2276
2277         for (i = 0; i < CONNTRACK_LOCKS; i++)
2278                 spin_lock_init(&nf_conntrack_locks[i]);
2279
2280         if (!nf_conntrack_htable_size) {
2281                 /* Idea from tcp.c: use 1/16384 of memory.
2282                  * On i386: 32MB machine has 512 buckets.
2283                  * >= 1GB machines have 16384 buckets.
2284                  * >= 4GB machines have 65536 buckets.
2285                  */
2286                 nf_conntrack_htable_size
2287                         = (((nr_pages << PAGE_SHIFT) / 16384)
2288                            / sizeof(struct hlist_head));
2289                 if (nr_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
2290                         nf_conntrack_htable_size = 65536;
2291                 else if (nr_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
2292                         nf_conntrack_htable_size = 16384;
2293                 if (nf_conntrack_htable_size < 32)
2294                         nf_conntrack_htable_size = 32;
2295
2296                 /* Use a max. factor of four by default to get the same max as
2297                  * with the old struct list_heads. When a table size is given
2298                  * we use the old value of 8 to avoid reducing the max.
2299                  * entries. */
2300                 max_factor = 4;
2301         }
2302
2303         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
2304         if (!nf_conntrack_hash)
2305                 return -ENOMEM;
2306
2307         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
2308
2309         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
2310                                                 sizeof(struct nf_conn),
2311                                                 NFCT_INFOMASK + 1,
2312                                                 SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
2313         if (!nf_conntrack_cachep)
2314                 goto err_cachep;
2315
2316         ret = nf_conntrack_expect_init();
2317         if (ret < 0)
2318                 goto err_expect;
2319
2320         ret = nf_conntrack_acct_init();
2321         if (ret < 0)
2322                 goto err_acct;
2323
2324         ret = nf_conntrack_tstamp_init();
2325         if (ret < 0)
2326                 goto err_tstamp;
2327
2328         ret = nf_conntrack_ecache_init();
2329         if (ret < 0)
2330                 goto err_ecache;
2331
2332         ret = nf_conntrack_timeout_init();
2333         if (ret < 0)
2334                 goto err_timeout;
2335
2336         ret = nf_conntrack_helper_init();
2337         if (ret < 0)
2338                 goto err_helper;
2339
2340         ret = nf_conntrack_labels_init();
2341         if (ret < 0)
2342                 goto err_labels;
2343
2344         ret = nf_conntrack_seqadj_init();
2345         if (ret < 0)
2346                 goto err_seqadj;
2347
2348         ret = nf_conntrack_proto_init();
2349         if (ret < 0)
2350                 goto err_proto;
2351
2352         conntrack_gc_work_init(&conntrack_gc_work);
2353         queue_delayed_work(system_power_efficient_wq, &conntrack_gc_work.dwork, HZ);
2354
2355         return 0;
2356
2357 err_proto:
2358         nf_conntrack_seqadj_fini();
2359 err_seqadj:
2360         nf_conntrack_labels_fini();
2361 err_labels:
2362         nf_conntrack_helper_fini();
2363 err_helper:
2364         nf_conntrack_timeout_fini();
2365 err_timeout:
2366         nf_conntrack_ecache_fini();
2367 err_ecache:
2368         nf_conntrack_tstamp_fini();
2369 err_tstamp:
2370         nf_conntrack_acct_fini();
2371 err_acct:
2372         nf_conntrack_expect_fini();
2373 err_expect:
2374         kmem_cache_destroy(nf_conntrack_cachep);
2375 err_cachep:
2376         kvfree(nf_conntrack_hash);
2377         return ret;
2378 }
2379
2380 static struct nf_ct_hook nf_conntrack_hook = {
2381         .update         = nf_conntrack_update,
2382         .destroy        = destroy_conntrack,
2383         .get_tuple_skb  = nf_conntrack_get_tuple_skb,
2384 };
2385
2386 void nf_conntrack_init_end(void)
2387 {
2388         /* For use by REJECT target */
2389         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
2390         RCU_INIT_POINTER(nf_ct_hook, &nf_conntrack_hook);
2391 }
2392
2393 /*
2394  * We need to use special "null" values, not used in hash table
2395  */
2396 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
2397 #define DYING_NULLS_VAL         ((1<<30)+1)
2398 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
2399
2400 int nf_conntrack_init_net(struct net *net)
2401 {
2402         int ret = -ENOMEM;
2403         int cpu;
2404
2405         BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER);
2406         atomic_set(&net->ct.count, 0);
2407
2408         net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
2409         if (!net->ct.pcpu_lists)
2410                 goto err_stat;
2411
2412         for_each_possible_cpu(cpu) {
2413                 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
2414
2415                 spin_lock_init(&pcpu->lock);
2416                 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
2417                 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
2418         }
2419
2420         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
2421         if (!net->ct.stat)
2422                 goto err_pcpu_lists;
2423
2424         ret = nf_conntrack_expect_pernet_init(net);
2425         if (ret < 0)
2426                 goto err_expect;
2427
2428         nf_conntrack_acct_pernet_init(net);
2429         nf_conntrack_tstamp_pernet_init(net);
2430         nf_conntrack_ecache_pernet_init(net);
2431         nf_conntrack_helper_pernet_init(net);
2432
2433         ret = nf_conntrack_proto_pernet_init(net);
2434         if (ret < 0)
2435                 goto err_proto;
2436         return 0;
2437
2438 err_proto:
2439         nf_conntrack_ecache_pernet_fini(net);
2440         nf_conntrack_expect_pernet_fini(net);
2441 err_expect:
2442         free_percpu(net->ct.stat);
2443 err_pcpu_lists:
2444         free_percpu(net->ct.pcpu_lists);
2445 err_stat:
2446         return ret;
2447 }